ArticlePDF Available

Abstract and Figures

Researchers are combining self-reports from mobile surveys with passive data collection using sensors and apps on smartphones increasingly more often. While smartphones are commonly used in some groups of individuals, smartphone penetration is significantly lower in other groups. In addition, different operating systems (OSs) limit how mobile data can be collected passively. These limitations cause concern about coverage error in studies targeting the general population. Based on data from the Panel Study Labour Market and Social Security (PASS), an annual probability-based mixed-mode survey on the labor market and poverty in Germany, we find that smart-phone ownership and ownership of smartphones with specific OSs are correlated with a number of sociodemographic and substantive variables.
Content may be subject to copyright.
Research Article
Coverage Error in Data
Collection Combining
Mobile Surveys With
Passive Measurement
Using Apps: Data From
a German National Survey
Florian Keusch
1
, Sebastian Ba
¨hr
2
,
Georg-Christoph Haas
1,2
, Frauke Kreuter
1,2,3
and Mark Trappmann
2,4
Abstract
Researchers are combining self-reports from mobile surveys with passive
data collection using sensors and apps on smartphones increasingly more
often. While smartphones are commonly used in some groups of individuals,
smartphone penetration is significantly lower in other groups. In addition,
different operating systems (OSs) limit how mobile data can be collected
passively. These limitations cause concern about coverage error in studies
targeting the general population. Based on data from the Panel Study Labour
Market and Social Security (PASS), an annual probability-based mixed-mode
survey on the labor market and poverty in Germany, we find that smart-
phone ownership and ownership of smartphones with specific OSs are
correlated with a number of sociodemographic and substantive variables.
1
University of Mannheim, Mannheim, Germany
2
Institute for Employment Research, Nuremberg, Germany
3
University of Maryland, College Park, MD, USA
4
University of Bamberg, Germany
Corresponding Author:
Florian Keusch, University of Mannheim, A5, 6, 68159 Mannheim, Germany.
Email: f.keusch@uni-mannheim.de
Sociological Methods & Research
1-38
ªThe Author(s) 2020
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/0049124120914924
journals.sagepub.com/home/smr
The use of weighting techniques based on sociodemographic information
available for both owners and nonowners reduces these differences but does
not eliminate them.
Keywords
smartphones, operating systems, coverage error, mobile web surveys,
passive mobile data collection
The advent of smartphones and the rapid expansion of high-speed cellular
Internet and Wi-Fi availability allow many people to use the Internet anytime
and anywhere. Researchers now can employ smartphones for data collection
in large populations using different methods, that is, for mobile web surveys
(e.g., Couper, Antoun, and Mavletova 2017; Keusch and Yan 2017), passive
mobile data collection of behavioral and digital traces via sensors and apps
(e.g., Keusch, Leonard, et al. 2019; Kreuter et al. 2018; Scherpenzeel 2017;
Sugie 2018), and data collection through other features of the smartphone
such as the built-in camera for taking pictures (e.g., Ja¨ckle et al. 2019; Yan
et al. 2017). However, while smartphone penetration is growing, the propor-
tion of smartphone owners in some subpopulations is lower than those in
others. Users further differ by the type of smartphone they use and its oper-
ating system (OS; i.e., Android, Apple iOS, Windows). OSs vary in the
amount and type of data that can be collected on a device, which is a
limitation for studies that use sensors and apps to passively measure user
behavior. For example, the architecture of the iOS and Windows systems
currently do not allow mobile data, such as geolocation, call and text logs,
and online browsing behavior, to be passively collected from smartphones
with the same level of detail as Android. Even for data that can be collected
on all OSs, developing a research app that runs on multiple OSs increases
study costs. Thus, it is not surprising that many studies rely on the use of only
a single OS (Church et al. 2015). If there are systematic differences between
smartphone users and nonusers or users of different OSs in the variable of
interest, coverage bias would arise in a study that uses smartphone technol-
ogy for data collection, meaning certain subgroups of the population are
systematically excluded from a sampling frame, and thus, no inference to
the general population can be drawn (Groves et al. 2009).
Studying coverage error is not trivial because little to no information
about the underrepresented population is available, and in some instances,
2Sociological Methods & Research XX(X)
it is difficult to separate coverage error from nonresponse error (see Eckman
and Kreuter [2017] for a recent summary). To study coverage error, one
needs auxiliary information (1) on who is and is not covered by a mode
or technology and (2) on the variables of interest for both the covered
and the noncovered population. To our knowledge, only a small number
of studies have explored coverage error when using smartphones for
mobile survey data collection (Antoun et al. 2018; Baier, Metzler, and
Fuchs 2018; Couper et al. 2018; Fuchs and Busse 2009; Metzler and
Fuchs 2014), in part because information on the population is rarely
available. To date, no study has specifically assessed the potential for
coverage bias when using a specific OS. The study presented in this
article is the first to use an administrative population frame and a
probability-based survey to assess coverage error and quantify the biases
in a group of variables on sociodemographics, behaviors, and attitudes
related to employment and poverty due to differences in smartphone
ownership and OSs.
Our article extends the existing research on smartphone coverage error
in several directions. We address the following specific research questions
in Germany:
1. What is the extent of smartphone coverage, Android smartphone
coverage, and iPhone coverage in the German residential population
aged 15 and older?
2. Does coverage differ for key sociodemographic subgroups?
3. To what extent does coverage affect key substantive outcome vari-
ables of the labor market and poverty survey?
4. Can coverage bias be reduced by implementing specific weights that
correct for known differences in device ownership?
We begin with a review of the existing literature on smartphone penetra-
tion and how the digital divide can lead to coverage error. Recent work has
focused on differences between smartphone users and nonusers. Little infor-
mation is known about whether differences between users of different OSs
lead to coverage error as well, and our article extends the research on cov-
erage error in this direction. We then describe the Panel Study Labour Market
and Social Security (PASS), an annual mixed-mode survey conducted by the
Institute for Employment Research in the German residential population
(Trappmann et al. 2019). We explore the potential for coverage error in
measures included in PASS when smartphone technology is used in this
population. Our study shows that smartphone ownership and ownership of
Keusch et al. 3
smartphones that operate on a specific OS correlate with sociodemographic
as well as substantive variables.
Smartphones and the Digital Divide
Smartphone Penetration
The popularity of mobile devices is reflected in an increase in smartphone
penetration in many countries over the last few years. Based on a national
telephone survey, Pew Research (2017) estimated that 77 percent of the U.S.
adult population owned a smartphone in 2016 and 46 percent owned a
smartphone in 2012. Eurostat (2018) reported that in 2011, only 19 percent
of Europeans used a smartphone (or another mobile phone) to access the
Internet. In 2017, this proportion increased to 63 percent. Mobile Internet
access varies considerably by country, and it ranged from 31 percent in Italy
to 84 percent in the Netherlands and Sweden in 2017. In Germany, the
proportion of people who used a smartphone (or another mobile phone) to
access the Internet increased from 15 percent in 2011 to 73 percent in 2017.
Despite the fact that mobile Internet access is increasing, smartphone
ownership exhibits large variability across subpopulations, leading to con-
cerns about coverage bias. In 2016, 92 percent of U.S. adults under the age of
30 owned a smartphone, whereas only 42 percent of adults 65 years and older
owned a smartphone. Smartphone ownership increases with household
income and educational attainment, and U.S. adults living in rural areas are
less likely to own a smartphone than are adults living in urban or suburban
areas (Pew Research 2017). Similarly, for Germany, Eurostat (2018) data
show that the proportion of people who use smartphones to access the Inter-
net decreases with age and increases with educational attainment, household
income, and population density. In addition, unemployed Germans are less
likely to go online with a smartphone than are employed people.
Using 2007 data from the Eurobarometer, a face-to-face interview survey,
Fuchs and Busse (2009) showed not only that smartphone penetration varies
across 33 European countries but also that substantial coverage bias toward
specific sociodemographic characteristics exists. Overall, the authors found a
higher likelihood of smartphone ownership among Europeans who are male,
younger, single, and live in nonrural areas and who were older when they
obtained their most recent educational degree. An update on this research
based on data from Eurobarometer 2013 (Metzler and Fuchs 2014) and 2017
(Baier et al. 2018) confirmed that mobile Internet access had increased in
Europe. This trend leads to a moderate reduction in bias toward specific
sociodemographic characteristics, and the level of coverage bias for mobile
4Sociological Methods & Research XX(X)
Internet converges toward that for landline Internet. In summary, studies
comparing smartphone owners and nonowners, including individuals who
own regular, nonsmartphone cell phones, confirm that a digital divide that is
driven by sociodemographic differences in age, education, income, and
employment status exists (Fortney et al. 2011; Park and Lee 2015).
While standard weighting procedures can account for differences in
observed sociodemographics between covered and uncovered members of
a population (Valliant, Dever, and Kreuter 2013), for users and nonusers of
smartphones, there are concerns that many unobserved variables such as
behaviors and attitudes will not be accounted for by weighting of socio-
demographics. To date, only a few studies have examined the effect of
smartphone coverage bias on substantive topics. First, Antoun et al. (2018)
used data from the probability-based Longitudinal Internet Studies for the
Social sciences (LISS) web panel in the Netherlands in 2013 to assess the
influence of smartphone ownership on coverage bias before and after con-
trolling for sociodemographic variables. Out of 19 substantive variables on
technology, lifestyle, and politics, eight showed significant coverage bias.
Two of these eight variables were associated with coverage even after socio-
demographic controls were applied, indicating that weighting techniques
based on sociodemographic information do not account for bias in these
substantive measures. Smartphone owners were more likely to prefer using
a tablet to go online and were more likely to report eating out in restaurants at
least once per month compared to those who did not own a smartphone. The
authors also found that coverage error was the largest contributor to total
survey error, compared to nonresponse and measurement error.
Second, Couper et al. (2018) estimated smartphone ownership in the
population of the National Survey of Family Growth (NSFG), a U.S. national
probability face-to-face survey with a repeated cross-sectional sample of
females and males aged 15–44. Based on NSFG data from 2012 to 2016,
81.6 percent of the study population had a smartphone. In terms of the socio-
demographic variables, the authors found that teenagers were less likely to
own a smartphone than were older groups of individuals, and married people
and people without children were more likely to own a smartphone than were
unmarried people and people with children, respectively. Confirming the
findings from Pew Research (2017), urbanicity, income, and educational
attainment were also significantly correlated with smartphone ownership.
In terms of substantive measures from the NSFG, the authors found few
consistent effects after controlling for demographics, and for most of the
demographic variables, the bias was relatively small. In particular, core
NSFG measures on divorce, cohabitation, and various dimensions of fertility
Keusch et al. 5
behaviors were not strongly associated with smartphone coverage. For atti-
tudinal questions, the bias introduced by smartphone coverage never
exceeded 2.5 percentage points.
Lane and Manner (2011) found in a volunteer sample of 312 cell phone
owners that smartphone owners reported higher levels of extroversion than
did owners of nonsmartphone cell phones, but the authors reported that there
were no differences between groups in the other four traits of the Big Five
personality model. Park and Lee (2015) surveyed 395 undergraduate students
and found that those owning a smartphone were more likely to be early
adopters of digital technologies, reported having a significantly higher num-
ber of online friends and online-only friends, and reported more social
engagement and political empowerment than nonowners.
Based on this evidence from previous studies, we expect to find a similar
digital divide between owners and nonowners of smartphones in a
population-based study on labor market activities and poverty in Germany,
particularly in measures related to age, education, income, and employment
status. In addition, we anticipate that coverage bias will extend to attitudinal
measures even after the models are adjusted for observed sociodemographic
differences between owners and nonowners.
Differences Between Owners of Smartphones Based on the OSs
Given that many smartphone studies run on only one OS (Church et al.
2015), it is worth investigating error that might arise from different types
of OSs, particularly given the large price differential usually associated with
the different OSs. We are not aware of any surveys that estimate the pene-
tration of smartphones with different OSs in the general population. The only
figures available on this topic are market shares. According to Kantar (2017),
83.2 percent of newly sold smartphones in Germany between August and
October 2017 were Android devices, and 14.9 percent were Apple iPhones.
However, these numbers reflect new sales only and thus do not necessarily
provide a good estimate for the total user population.
Currently, there are no studies based on probability samples that have
estimated coverage bias that stems from owning a smartphone with a specific
OS. Several studies used volunteer samples to compare the characteristics of
owners of different smartphones, and the results indicated that individuals
who own Apple iPhones and those who own Android smartphones are indeed
different groups of people. In terms of sociodemographic characteristics,
several studies found that iPhone owners are younger (Pryss et al. 2018;
6Sociological Methods & Research XX(X)
Shaw et al. 2016) and more likely to be female than are Android smartphone
owners (Go¨tz, Stieger, and Reips 2017; Shaw et al. 2016; Ubhi et al. 2017).
Another consistent finding is that owners of smartphones with different
OSs seem to differ in terms of their financial resources. iPhone owners
reported having larger monthly budgets available (Go¨tz et al. 2017) and
spending more money on various categories of monthly expenses such as
clothing, makeup and beauty products, and tech-related purchases compared
to Android smartphone owners (Schmall 2018; Yarrow 2013). In contrast,
Android smartphone owners are more likely to consider themselves a frugal
person and to look for deals and discounts often (Schmall 2018). These
findings are consistent with the price difference between Android smart-
phones and iPhones.
In addition, some studies have found differences in personality and atti-
tudes of smartphone owners depending on the smartphone OS (see Go¨tzetal.
[2017] for an exception). For example, Shaw et al. (2016) reported that
Android users displayed higher levels of honesty-humility, openness, and
avoidance similarity, and they scored lower in emotionality than did iPhone
owners. Schmall (2018) reported of a study that found that iPhone owners
more often state that they like being the center of attention and are more
likely to be very happy with their life in general and their current job in
particular. In the same study, iPhone owners considered themselves more
personable and sociable than Android users, and they reported having more
close friends and being more active and social on weekends.
Finally, Shaw et al. (2016) found that Android smartphone owners felt
that their smartphone was less of a status object than iPhone owners did. In
summary, these findings on differences in the personalities of smartphone
owners are in line with the personality of the Apple brand, which focuses on
nonconformity, innovation, and creativity as well as belonging to a social
elite (Dissanayake and Amarasuriya 2015; Go¨tz et al. 2017).
Based on these findings in the literature, we expect to see differences
between people who own smartphones that run on different OSs, especially
in age and socioeconomic status, as well as in attitudinal measures, even after
controlling for sociodemographic differences. We used data from the PASS in
Germany, which is described in the next section, to study coverage error when
smartphones are used for collecting data on labor market activities and poverty.
PASS
PASS is a yearly household panel survey conducted in the residential pop-
ulation in Germany (Trappmann et al. 2019). The main purpose of PASS is to
Keusch et al. 7
create a data source for evaluating labor market and welfare state reforms and
for enabling longitudinal research on the labor market and on poverty in
Germany. The initial residential population sample was drawn from an
address register of a commercial supplier (Rudolph and Trappmann 2007)
with refreshments drawn from municipal registers before waves 5 and 11.
Welfare benefit recipient households are oversampled based on national
benefit recipient registers. PASS is conducted as a mixed mode survey utiliz-
ing Computer-Assisted Personal Interviewing (CAPI) as the main mode and
Computer-Assisted Telephone Interviewing (CATI) as a follow-up mode for
initial nonresponders.
1
However, the mode that was successful in the previ-
ous wave becomes the mode that is initially used in the following wave
(Trappmann, Mu
¨ller, and Bethmann 2013).
Each head of a household is asked to participated in a 15-minute
household-related interview about topics such as household composition,
dwelling, childcare, benefit receipt, household income, and deprivation. Sub-
sequently, every household member aged 15 and older is asked to complete a
30-minute person-related interview on topics such as job searching and labor
market participation, job quality, social inclusion, attitudes, and personality
traits (Beste et al. 2013).
PASS provides a set of weights that allows researchers to weight all
interviewed households and persons in a given wave to all households and
persons in Germany at that time (Trappmann 2013). Detailed information on
sampling and weighting can be found in the yearly data reports (for Wave 11,
see Berg et al. 2018) and methods reports (for Wave 11, see Jesske and
Schulz 2018). For the remainder of this article, the “standard PASS weights”
always correspond to these PASS weights.
2
Methods and Data
Between February and September 2017, 13,703 individuals in 9,420 house-
holds were interviewed as part of PASS Wave 11. The personal questionnaire
included two questions about smartphone ownership (see Figure 1).
Based on these two questions, we assess coverage error in the German
residential population aged 15 and older regarding (1) smartphone owner-
ship, (2) Android smartphone ownership, and (3) iPhone ownership, first
with an overall estimate of smartphone ownership and ownership of different
OSs. We then consider (a) key sociodemographic variables that are assessed
in previous research on smartphone coverage (Antoun et al. 2018; Baier et al.
2018; Couper et al. 2018; Fuchs and Busse 2009; Metzler and Fuchs 2014;
Pew Research 2017) and are traditionally used when calculating weights for
8Sociological Methods & Research XX(X)
data from general population surveys. For Germany and the PASS, these
sociodemographic weighting variables include age, sex, educational attain-
ment, nationality, region, and community size. For each of these sociodemo-
graphic weighting variables, we assess differences in smartphone ownership,
Android smartphone ownership, and iPhone ownership. Next, we consider
substantive measures collected as part of PASS. We assess (b) other socio-
demographic variables collected as part of PASS that are usually not avail-
able for weighting in Germany, such as marital status, household size, and
the prevalence of individuals who live with their own children in the house-
hold. We then consider (c) substantive measures related to employment and
poverty, that is, household income, employment status, working hours, wel-
fare receipt, and deprivation.
3
Finally, we assess (d) attitudinal measures on
different aspects of life satisfaction
4
(i.e., life in general, health, housing,
living standards) and self-efficacy
5
as well as (e) measures of social embedd-
edness, that is, personal network size and social inclusion.
6
Table 1 presents
all variables used in our analysis and how they were operationalized. The
exact wording for the original items can be found at http://fdz.iab.de/en/
FDZ_Individual_Data/PASS.aspx.
To assess the influence of smartphone, Android smartphone, and iPhone
coverage on the substantive PASS measures, we produce estimates for the 15
substantive PASS variables (variable groups [b] through [e] in Table 1). We
produce the estimates by following multiple steps: First, we estimate these
outcomes based on the full PASS sample using the standard PASS weights
(see Trappmann 2013). This procedure provides unbiased population esti-
mates of the core outcomes of the PASS. Second, we estimate the same
outcomes based on all smartphone owners, all Android smartphone owners,
P 298. Do you own a mobile phone with Internet access, a so called smartphone?
1 YES
2NO
[IF P298 = 1]
P299. Which of the following operating systems is installed on your smartphone?
1 Android
2 Apple iOS
3 Windows
4 Another operating system, please specify
Figure 1. Questions on smartphone ownership and operating system in Panel Study
Labour Market and Social Security Wave 11 (original questions are in German).
Keusch et al. 9
Table 1. Variables Used to Assess Bias Due to Smartphone, Android Smartphone,
and iPhone Coverage in Panel Study Labour Market and Social Security Wave 11.
Variable Unweighted (n) Unweighted (%)
Outcome variables
Smartphone ownership
Yes 10,677 78.0
Missing 10
Operating system of smartphone
Android 7,255 72.1
Apple iOS 2,018 20.1
Windows 706 7.0
Other 80 0.8
Don’t know 584
Missing 3,060
(a) Sociodemographic weighting variables
Sex
Male 6,786 49.5
Female 6,917 50.5
Missing 0
Age
15–19 783 5.7
20–24 926 6.8
25–29 1,168 8.5
30–34 1,310 9.6
35–39 1,131 8.3
40–44 962 7.0
45–49 1,211 8.8
50–54 1,424 10.4
55–59 1,505 11.0
60–64 1,309 9.6
65–69 1,015 7.4
70–74 433 3.2
75þ526 3.8
Missing 0
Education
Primary and secondary I 2,503 18.3
Secondary II 5,883 43.0
Postsecondary 4,809 35.2
Still in education 486 3.6
Missing 22
(continued)
10 Sociological Methods & Research XX(X)
Table 1. (continued)
Variable Unweighted (n) Unweighted (%)
Nationality
German 11,478 83.9
Non-German 2,211 16.2
Missing 14
Region
Old states (former West Germany) 10,045 73.3
New states (former East Germany) 3,658 26.7
Missing 0
Community size (No. of inhabitants)
Less than 5,000 466 3.4
5,000–19,999 1,474 10.8
20,000–49,999 1,413 10.3
50,0000–99,999 1,489 10.9
100,000–499,999 4,166 30.4
500,000þ4,695 34.3
Missing 0
(b) Other sociodemographic variables
Marital status
Married 6,087 44.9
Divorced, widowed, separated 2,765 20.4
Never married 4,700 34.7
Missing 151
Household (HH) size
1 3,462 25.3
2 4,661 34.0
3 2,367 17.3
4 1,883 13.7
5 or more 1,330 9.7
Missing 0
Own children in HH
Yes 4,849 35.4
Missing 30
(c) Variables on employment and poverty
HH income
Less than 1,000 euros 4,657 34.3
1,000–1,999 euros 6,027 44.4
2,000 euros or more 2,877 21.2
Missing 142
Employment status
Employed 6,030 44.3
Unemployed 2,608 19.2
(continued)
Keusch et al. 11
Table 1. (continued)
Variable Unweighted (n) Unweighted (%)
Inactive 4,979 36.6
Missing 86
Working hours
Under 35 hours 1,808 31.5
35–40 hours 2,017 35.2
More than 40 hours 1,906 33.3
Missing 7,972
Welfare receipt
Yes 3,367 24.6
Missing 41
Deprivation
a
No deprivation 5,772 42.1
Medium deprivation 4,240 30.9
High deprivation 3,691 26.9
Missing 0
(d) Attitudinal variables
Satisfaction with life in general
b
Low 670 4.9
Medium 6,154 55.0
High 6,860 50.1
Missing 19
Satisfaction with health
b
Low 1,240 9.1
Medium 6,188 45.2
High 6,270 45.7
Missing 5
Satisfaction with housing
b
Low 670 4.9
Medium 4,438 32.4
High 8,588 62.7
Missing 7
Satisfaction with living standards
b
Low 840 6.1
Medium 5,900 43.1
High 6,948 50.8
Missing 15
Self-efficacy
c
Low 4,582 33.9
Medium 5,557 41.2
(continued)
12 Sociological Methods & Research XX(X)
and all iPhone owners in our sample, again using the standard PASS weights.
This provides estimates that we would obtain if we collected these measures
based on samples of smartphone owners or owners of a smartphone with a
specific OS, without specifically correcting for sociodemographic differ-
ences due to smartphone ownership or ownership of smartphones with a
specific OS. Finally, to account for the fact that smartphone ownership and
ownership of Android smartphones and iPhones varies substantially by
sociodemographic variables, we estimate the 15 substantial PASS outcomes
based on all smartphone owners, all Android smartphone owners, and all
iPhone owners in our sample, but this time we adjust the standard PASS
weights. To this end, we specify three models predicting smartphone own-
ership and ownership of a smartphone with a specific OS: one model with a
binary dependent variable for owning a smartphone (vs. not owning a smart-
phone), one model with a binary dependent variable for owning an Android
smartphone (vs. owning a smartphone with a different OS or not owning a
smartphone at all), and one model with a binary dependent variable for
Table 1. (continued)
Variable Unweighted (n) Unweighted (%)
High 3,358 24.9
Missing 206
(e) Variables on social embeddedness
Size of personal network
0–2 2,352 17.3
3–9 7,369 54.1
10 or more 3,899 28.6
Missing 83
Social inclusion
d
Low 933 6.9
Medium 5,366 39.4
High 7,305 53.7
Missing 99
Note:N¼13,703.
a
Deprivation is measured as an index using 23 items available in households weighted based on
needfulness ranging from 0 to 8.2. No deprivation (all items available) ¼0, medium deprivation ¼
0.1–0.9, and high deprivation ¼1.0–8.2.
b
Original satisfaction scale ranged from 0 ¼completely unsatisfied to 10 ¼completely satisfied.
Low ¼0–3, medium ¼4–7, and high ¼8–10.
c
Based on composite score from 5 items.
d
Original social inclusion scale ranged from 1 ¼excluded to 10 ¼included. Low ¼1–3, medium ¼
4–7, and high ¼8–10.
Keusch et al. 13
owning an iPhone (vs. owning a smartphone with a different OS or not
owning a smartphone at all). Each of the models uses the six sociodemo-
graphic weighting variables (i.e., age, sex, educational attainment, national-
ity, region, and community size) as predictors. We then create deciles for the
predicted probabilities, and we multiply the inverse of the decile group
means with the standard PASS weights. The product of this operation is
what we refer to as the “adjusted PASS weights.”
To determine the magnitude of coverage bias in the substantive PASS
variables, we follow the same approach as Couper et al. (2018) and estimate
the difference between the full sample (f) and the sample covered (c)by
smartphones, Android smartphones, and iPhones as
biasðyÞ¼ycyf:ð1Þ
Next, we calculate the standard error of the estimated bias as
SEðycyfÞ¼nfnc
nf
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
varðycÞþvarðyncÞ
p:ð2Þ
To test the significance of a given bias, we use a z-test. To calculate a test
statistic, we divide the estimate of the difference by the standard error of the
difference.
For all our analyses, we report estimates based on complete data and use case-
wise deletion for multivariate models. We use Stata 15.1 with the svy command
when applying weights in our analysis, thereby accounting for clustering and
stratification of the survey design when estimating the standard errors.
Results
What Is the Extent of Smartphone Coverage, Android Smartphone
Coverage, and iPhone Coverage in the German Residential Population
Aged 15 and Older?
Based on the weighted estimates from PASS Wave 11, 75.8 percent of
German residents aged 15 and older own a smartphone. Regarding the
specific OS, 49.0 percent of German residents aged 15 and older own a
smartphone with the Android OS, 16.7 own an iPhone, 5.4 own a Windows
smartphone, and 0.7 percent own a smartphone that runs on another OS.
Approximately 4 percent do not know on which OS their smartphone is
operating. Due to the extremely small penetration of smartphones operating
on platforms other than Android and iOS, we did not consider them indi-
vidually for further analyses.
14 Sociological Methods & Research XX(X)
Does Coverage Differ for Key Sociodemographic Subgroups?
Table 2 presents weighted estimates (using standard PASS weights) of cov-
erage rates for all smartphone owners, Android smartphone owners, and
iPhone owners in Germany by key sociodemographic characteristics. Here,
we include only variables for which information is available on the general
population in Germany and that can thus be used for weighting
(“sociodemographic weighting variables”). We test for significant differences
in these characteristics using design-adjusted Rao-Scott chi-square tests.
From Table 2, we see that smartphone ownership and iPhone ownership do
significantly differ by sex; 78.7 percent of male German residents aged 15 and
older reported that they own a smartphone, while only 73.0 percent of female
German residents aged 15 and older reported that they own a smartphone.
Males (52.8 percent) are also significantly more likely to own an Android
smartphone than are females (45.3 percent), while there is no significant
difference between genders in iPhone ownership. Regarding age, while smart-
phone ownership is relatively consistent and remains above 95 percent for
German residents under the age of 45, the likelihood of smartphone ownership
drops significantly and substantially for older age groups. For example, fewer
than 5 out of 10 German residents aged 65–69 and only approximately only 2
out of 10 aged 75 and older own a smartphone. While a similarly sharp
decrease in ownership for older age groups can be observed for Android users,
Android smartphone ownership peaks between ages 25 and 39 at approxi-
mately 70 percent. However, teenagers and young adults under the age of
25 are overrepresented among iPhone owners, with approximately 30 percent
iPhone ownership in this age-group. Figure 2 shows the relationship between
age groups and smartphone ownership as well as between age groups and OS
ownership based on a locally weighted scatter-plot smoother regressions.
Smartphone, Android smartphone, and iPhone ownership significantly
increase with the level of education; 81.8 percent of the German residential
population with a postsecondary degree own a smartphone, 51.8 percent own
an Android smartphone, and 20.9 percent own an iPhone, while less than 70,
44, and 14 percent of German residents with a low level of formal education
own a smartphone, an Android smartphone, and an iPhone, respectively.
Smartphone, Android smartphone, and iPhone ownership are highest among
individuals currently pursuing a degree. Given that people pursuing a degree
tend to be younger, this result is not surprising.
Smartphone ownership is also significantly associated with nationality;
the proportion of German citizens who are smartphone owners (74.8 percent)
is smaller than that of non-German citizens (84.7 percent). However, the
Keusch et al. 15
Table 2. Weighted Estimates
a
, Standard Errors, and p-Values From Rao-Scott
Chi-Square Tests Comparing Smartphone, Android Smartphone, and iPhone Cover-
age Rates by Sociodemographic Indicators From Panel Study Labour Market and
Social Security (PASS) Wave 11.
Variable
Unweighted
(n)
Weighted
(N)
Smartphone Android iPhone
%SE %SE %SE
Overall 13,703 70,519,000 75.77 0.87 49.01 1.03 16.71 0.80
Sex p< .001 p< .001 p>.05
Male 6,786 34,673,000 78.68 1.01 52.84 1.33 17.34 1.04
Female 6,917 35,846,000 72.96 1.15 45.30 1.23 16.09 0.80
Age p< .001 p< .001 p< .001
15–19 783 4,109,000 97.14 1.24 66.47 3.36 29.91 3.39
20–24 926 4,390,000 98.39 0.75 62.71 3.60 32.92 3.55
25–29 1,168 5,336,000 98.04 0.73 71.08 3.24 21.87 3.12
30–34 1,310 5,218,000 95.54 1.23 68.39 3.12 21.18 2.82
35–39 1,131 5,037,000 95.43 1.15 70.89 3.24 17.46 3.05
40–44 962 5,022,000 95.44 1.26 61.49 3.75 23.79 3.17
45–49 1,211 6,341,000 86.87 2.30 57.14 3.04 19.90 2.46
50–54 1,424 6,965,000 84.17 1.66 53.31 2.86 15.91 2.14
55–59 1,505 6,090,000 73.71 2.22 48.84 2.74 13.87 2.11
60–64 1,309 5,263,000 65.05 2.87 35.94 2.66 12.37 2.10
65–69 1,015 4,475,000 55.57 3.17 30.38 3.06 9.63 1.65
70–74 433 3,783,000 46.08 3.62 22.88 3.08 5.32 1.38
75þ526 8,490,000 20.10 2.47 7.41 1.47 3.10 0.97
Education p< .001 p< .001 p< .001
Primary and secondary I 2,503 10,428,614 67.97 2.65 43.59 2.60 13.57 1.81
Secondary II 5,883 30,934,106 71.37 1.42 46.34 1.45 13.65 1.08
Postsecondary 4,809 26,411,516 81.77 1.03 51.77 1.50 20.87 1.29
Still in education 486 2,581,191 98.07 0.94 73.07 4.07 24.48 4.06
Nationality
b
p< .001 p>.05 p>.05
German 11,478 62,670,895 74.78 0.93 48.57 1.02 16.24 0.83
Non-German 2,211 7,837,000 83.71 2.13 52.34 3.10 20.47 2.52
Region p< .001 p>.05 p¼.021
Old states (former West
Germany)
10,045 56,618,000 77.19 0.97 49.58 1.20 17.63 0.92
New states (former East
Germany)
3,658 13,901,000 69.95 1.91 46.69 1.78 12.97 1.61
Community size
(No. of inhabitants)
p¼.018 p>.05 p>.05
Less than 5,000 466 2,489,941 74.99 5.68 49.24 6.12 13.37 5.46
5,000–19,999 1,474 7,469,041 72.59 2.74 47.53 2.75 13.05 1.41
20,000–49,999 1,413 7,260,826 67.79 3.25 44.22 3.01 13.00 2.18
50,0000–99,999 1,489 6,053,095 76.23 3.35 51.40 3.37 12.56 1.71
100,000–499,999 4,166 21,326,259 79.46 1.27 52.58 1.90 17.48 1.52
500,000þ4,695 25,919,837 75.84 1.34 47.26 1.66 19.45 1.47
a
Weighted to German residential population aged 15 and older using PASS weights.
b
There are 14 missing values for nationality in PASS Wave 11.
16 Sociological Methods & Research XX(X)
association between Android smartphone and iPhone ownership and nation-
ality is not significant (p> .05).
Smartphone ownership also significantly differs by region. While 77.2
percent of German residents aged 15 and older living in one of the “Alte
Bundesla¨nder” (Old States) of former West Germany own a smartphone,
only 70.0 percent of people living in one of the “Neue Bundesla¨nder” (New
States) of former East Germany own a smartphone. A similar trend can be
observed for iPhone ownership (West: 17.6 percent, East: 13.0 percent), but
the association between region and Android smartphone ownership is not
statistically significant (p> .05).
Finally, smartphone ownership in Germany increases with community
size. In cities with a population of more than 50,000, the proportion of
smartphone owners is higher than 75 percent, while in smaller communities,
smartphone ownership is significantly lower. A similar trend can be observed
for iPhone ownership, but this association is not statistically significant
(p> .05). For Android ownership, there seems to be no clear pattern in
ownership by community size (p> .05).
To What Extent Does Coverage Affect Key Substantive Outcome
Variables of a Labor Market and Poverty Survey?
Tables 3–5 show the estimates and standard errors for the 15 substantive
PASS variables corresponding to respondents who reported owning a
Figure 2. Smartphone ownership and operating system ownership by age groups.
Keusch et al. 17
smartphone (Table 3), owning an Android smartphone (Table 4), and owning
an iPhone (Table 5). In each of the three tables, Column 2 presents the
unbiased estimates and standard errors from the full PASS sample, and
Column 3 presents the estimates and standard errors from the reduced sample
covered by the device applying the standard PASS weights. Column 5 pro-
vides the bias estimates and their standard errors when the full PASS sample
estimates are compared with the covered sample estimates using the standard
PASS weights. Estimates that are statistically significant based on the z-test
are in bold font.
Column 5 in Table 3 shows that the majority of core substantive PASS
variables are associated with smartphone coverage bias. Using the sample of
smartphone owners in the PASS, we underestimate the percentage of Ger-
man residents aged 15 and older who are divorced, widowed, or separated by
more than five percetage points (p.p.), and we overestimate the percentage of
people who never married by approximately the same amount. We also see
that people living in households with more than two people are significantly
overestimated and people living in single- and two-person households are
underestimated. Similarly, there is a positive coverage bias for people living
with their own children in the household (þ6.5 p.p.).
Regarding measures on employment and poverty, we find that the
smartphone-only sample overestimates the percentage of people living in
the highest household income bracket (2,000 euros or more) by almost four
points and underestimates the percentage of people living in the two lower
household income groups (less than 1,000 euros: 0.9 p.p.; 1,000–1,999
euros: 2.8 p.p.). Coverage bias seems especially strong for employment
status, where the smartphone-only sample overestimates the percentage of
employed German residents by nine points, while it underestimates the per-
centage of people inactive by the same amount. In contrast, estimates on
working hours, welfare receipt, and self-reported deprivation seem to be only
minimally affected by smartphone coverage.
In terms of attitudinal variables, we find relatively small coverage bias in
the measures of general life satisfaction (high: þ1.9 p.p.; medium: 0.7 p.p.;
low: 1.3 p.p.), satisfaction with housing (high: 1.1 p.p.; medium: þ0.8
p.p.; low: þ0.3 p.p.), and self-efficacy (high: ns.; medium: þ1.9 p.p.; low:
1.7 p.p.). A substantially larger bias can be observed in the estimates of
satisfaction with health; the sample covered by smartphones overestimates
the proportion of people reporting high satisfaction with their health by
almost five points (low: 1.7 p.p.; medium: 3.1 p.p.).
We also find only small biases due to smartphone ownership in the mea-
sures of social embeddedness; using the covered sample of smartphone
18 Sociological Methods & Research XX(X)
Table 3. Weighted Estimates and Standard Errors (SE) for Substantive Panel Study Labour Market and Social Security (PASS) Wave 11
Variables Based on Smartphone Ownership and Weighted Estimates of Smartphone Coverage Bias.
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of Smartphone Owners Smartphone Coverage Bias
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Marital status
Married 6,087 57.4 (1.1) 57.9 (1.2) 60.4 (1.2) 0.6 (.6) 3.0 (.6)
Divorced, widowed, separated 2,765 20.7 (1.0) 15.2 (0.9) 17.8 (1.0) 5.5 (.5) 3.0 (.6)
Never married 4,700 21.9 (0.7) 26.8 (0.9) 21.9 (0.8) 5.0 (.3) 0.0 (.2)
Household (HH) size
1 3,462 23.2 (0.9) 19.2 (0.9) 19.8 (0.9) 4.0 (.5) 3.5 (.6)
2 4,661 38.3 (1.1) 33.9 (1.2) 39.9 (1.4) 4.4 (.5) 1.6 (.6)
3 2,367 17.4 (0.9) 20.9 (1.1) 18.3 (1.0) 3.5 (.3) 0.9 (.3)
4 1,883 14.8 (0.9) 18.4 (1.1) 15.6 (0.9) 3.6 (.3) 0.8 (.3)
5 or more 1,330 6.3 (0.6) 7.7 (0.8) 6.5 (0.7) 1.3 (.2) 0.2 (.2)
Own children in HH
Yes 4,849 32.2 (1.0) 38.7 (1.1) 33.9 (1.1) 6.5 (.4) 1.7 (.3)
HH income
Less than 1,000 euros 4,657 15.9 (0.8) 15.1 (0.9) 14.4 (0.9) 0.9 (.4) 1.6 (.4)
1,000–1,999 euros 6,027 49.6 (1.3) 46.8 (1.5) 47.7 (1.5) 2.8 (.6) 1.9 (.6)
2,000 euros or more 2,877 34.4 (1.2) 38.1 (1.3) 38.0 (1.3) 3.7 (.5) 3.5 (.5)
Employment status
Employed 6,030 51.7 (0.9) 60.7 (0.9) 55.2 (1.1) 9.0 (.4) 3.5 (.4)
Unemployed 2,608 4.9 (0.3) 5.3 (0.4) 4.7 (0.3) 0.4 (.1) 0.3 (.1)
Inactive 4,979 43.3 (0.9) 34.0 (0.9) 40.1 (1.1) 9.4 (.4) 3.2 (.4)
(continued)
19
Table 3. (continued)
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of Smartphone Owners Smartphone Coverage Bias
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Working hours
Under 35 hours 1,808 27.8 (1.0) 27.0 (1.0) 27.8 (1.1) 0.8 (.4) 0.0 (.4)
35–40 hours 2,017 34.8 (1.3) 35.1 (1.3) 35.1 (1.3) 0.2 (.5) 0.2 (.5)
More than 40 hours 1,906 37.3 (1.3) 37.9 (1.4) 37.1 (1.3) 0.6 (.4) 0.2 (.4)
Welfare receipt
Yes 3,367 5.6 (0.4) 5.9 (0.4) 5.2 (0.4) 0.3 (.2) 0.3 (.1)
Deprivation
a
No deprivation 5,772 64.8 (1.2) 64.5 (1.3) 65.6 (1.3) 0.3 (.5) 0.8 (.6)
Medium deprivation 4,240 24.8 (1.0) 25.2 (1.2) 25.1 (1.3) 0.4 (.5) 0.3 (.5)
High deprivation 3,691 10.4 (0.7) 10.3 (0.8) 9.4 (0.7) 0.1 (.3) 1.0 (.3)
Satisfaction with life in general
b
Low 670 2.3 (0.3) 1.6 (0.2) 1.6 (0.2) 0.7 (.2) 0.7 (.2)
Medium 6,154 38.7 (1.1) 37.4 (1.2) 37.3 (1.3) 1.3 (.5) 1.4 (.6)
High 6,860 59.1 (1.2) 61.0 (1.2) 61.1 (1.3) 1.9 (.5) 2.0 (.6)
Satisfaction with health
b
Low 1,240 6.3 (0.4) 4.6 (0.4) 5.0 (0.5) 1.7 (.3) 1.3 (.3)
Medium 6,188 45.0 (1.0) 41.9 (1.1) 44.3 (1.1) 3.1 (.5) 0.7 (.6)
High 6,270 48.6 (1.0) 53.5 (1.1) 50.7 (1.1) 4.9 (.5) 2.0 (.5)
Satisfaction with housing
b
Low 670 2.4 (0.3) 2.7 (0.4) 2.6 (0.4) 0.3 (.1) 0.2 (.1)
Medium 4,438 26.2 (1.0) 27.0 (1.1) 25.9 (1.1) 0.8 (.4) 0.4 (.4)
High 8,588 71.4 (1.0) 70.3 (1.1) 71.6 (1.1) 1.1 (.4) 0.2 (.5)
(continued)
20
Table 3. (continued)
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of Smartphone Owners Smartphone Coverage Bias
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Satisfaction with living standards
b
Low 840 2.7 (0.3) 2.2 (0.3) 2.1 (0.3) 0.5 (.2) 0.6 (.2)
Medium 5,900 32.1 (1.1) 32.1 (1.2) 31.2 (1.1) 0.0 (.5) 0.9 (.5)
High 6,948 65.2 (1.1) 65.7 (1.2) 66.7 (1.2) 0.5 (.5) 1.5 (.6)
Self-efficacy
c
Low 4,582 28.7 (0.9) 27.0 (1.0) 26.6 (1.0) 1.7 (.4) 2.1 (.5)
Medium 5,557 46.8 (1.0) 48.7 (1.2) 47.9 (1.2) 1.9 (.5) 1.1 (.5)
High 3,358 24.5 (0.8) 24.3 (0.9) 25.5 (1.0) 0.2 (.4) 1.0 (.5)
Size of personal network
0–2 2,352 13.0 (0.7) 12.5 (0.8) 12.3 (0.9) 0.5 (.3) 0.7 (.4)
3–9 7,369 56.8 (0.9) 58.5 (1.1) 56.9 (1.2) 1.6 (.5) 0.1 (.6)
10 or more 3,899 30.0 (1.0) 29.1 (1.1) 30.7 (1.2) 1.1 (.5) 0.6 (.6)
Social inclusion
d
Low 933 3.4 (0.4) 2.2 (0.3) 2.3 (0.3) 1.2 (.2) 1.2 (.3)
Medium 5,366 34.0 (1.0) 32.7 (1.1) 32.3 (1.2) 1.2 (.5) 1.6 (.5)
High 7,305 62.6 (1.0) 65.0 (1.2) 65.4 (1.2) 2.4 (.5) 2.8 (.5)
Note: Bold indicates statistically significant bias estimates based on a z-test (p< .05).
a
Deprivation is measured as an index using 23 items available in households weighted based on needfulness ranging from 0 to 8.2. No deprivation (all items
available) ¼0, medium deprivation ¼0.1–0.9, and high deprivation ¼1.0–8.2.
b
Original satisfaction scale ranged from 0 ¼completely unsatisfied to 10 ¼completely satisfied. Low ¼0–3, medium ¼4–7, and high ¼8–10.
c
Based on composite score from 5 items.
d
Original social inclusion scale ranged from 1 ¼excluded to 10 ¼included. Low ¼1–3, medium ¼4–7, and high ¼8–10.
21
Table 4. Weighted Estimates and Standard Errors (SE) for Substantive Panel Study Labour Market and Social Security (PASS) Wave 11
Variables Based on Android Smartphone Ownership and Weighted Estimates of Android Smartphone Coverage Bias.
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of Android Smartphone Owners Android Smartphone Coverage Bias
With Standard
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Marital status
Married 6,087 57.4 (1.1) 58.5 (1.4) 60.9 (1.4) 1.2 (1.0) 3.5 (1.1)
Divorced, widowed, separated 2,765 20.7 (1.0) 13.8 (0.9) 15.7 (1.0) 6.9 (0.8) 5.1 (1.0)
Never married 4,700 21.9 (0.7) 27.6 (1.1) 23.4 (1.0) 5.7 (0.7) 1.6 (0.6)
Household (HH) size
1 3,462 23.2 (0.9) 18.1 (1.0) 18.2 (1.0) 5.2 (0.8) 5.1 (0.9)
2 4,661 38.3 (1.1) 32.4 (1.5) 38.0 (1.7) 5.8 (1.0) 0,3 (1.2)
3 2,367 17.4 (0.9) 22.4 (1.5) 20.1 (1.4) 5.0 (0.8) 2.8 (0.7)
4 1,883 14.8 (0.9) 19.0 (1.3) 16.6 (1.2) 4.2 (0.9) 1.9 (0.6)
5 or more 1,330 6.3 (0.6) 8.1 (0.9) 7.1 (0.8) 1.8 (0.5) 0.8 (0.4)
Own children in HH
Yes 4,849 32.2 (1.0) 41.1 (1.5) 37.1 (1.5) 8.9 (1.0) 4.9 (0.9)
HH income
Less than 1,000 euros 4,657 15.9 (0.8) 15.9 (1.7) 14.9 (1.0) 0.1 (0.7) 1.1 (0.8)
1,000–1,999 euros 6,027 49.6 (1.3) 48.3 (1.7) 48.9 (1.7) 1.3 (1.1) 0.7 (1.1)
2,000 euros or more 2,877 34.4 (1.2) 35.8 (1.6) 36.2 (1.5) 1.4 (1.1) 1.8 (1.1)
Employment status
Employed 6,030 51.7 (0.9) 62.1 (1.1) 57.8 (1.3) 10.4 (0.8) 6.2 (0.9)
Unemployed 2,608 4.9 (0.3) 6.0 (0.5) 5.4 (0.4) 1.0 (0.3) 0.5 (0.2)
Inactive 4,979 43.3 (0.9) 32.0 (1.1) 36.8 (1.3) 11.4 (0.8) 6.6 (0.9)
(continued)
22
Table 4. (continued)
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of Android Smartphone Owners Android Smartphone Coverage Bias
With Standard
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Working hours
Under 35 hours 1,808 27.8 (1.0) 26.2 (1.2) 26.9 (1.2) 1.7 (0.9) 0.9 (0.9)
35–40 hours 2,017 34.8 (1.3) 37.3 (1.6) 37.2 (1.6) 2.4 (1.0) 2.3 (1.0)
More than 40 hours 1,906 37.3 (1.3) 36.6 (1.6) 36.0 (1.6) 0.8 (1.0) 1.4 (1.0)
Welfare receipt
Yes 3,367 5.6 (0.4) 6.4 (0.5) 5.9 (0.4) 0.9 (0.3) 0.4 (0.3)
Deprivation
a
No deprivation 5,772 64.8 (1.2) 63.4 (1.5) 64.6 (1.5) 1.4 (1.0) 0.2 (1.0)
Medium deprivation 4,240 24.8 (1.0) 25.1 (1.3) 24.8 (1.3) 0.3 (0.9) 0.0 (0.9)
High deprivation 3,691 10.4 (0.7) 11.5 (1.0) 10.5 (0.9) 1.1 (0.6) 0.2 (0.6)
Satisfaction with life in general
b
Low 670 2.3 (0.3) 1.4 (0.2) 1.4 (0.2) 0.8 (0.3) 1.3 (0.4)
Medium 6,154 38.7 (1.1) 38.3 (1.5) 38.5 (1.6) 0.4 (1.0) 0.3 (0.9)
High 6,860 59.1 (1.2) 60.3 (1.5) 60.1 (1.6) 1.2 (1.0) 1.0 (1.1)
Satisfaction with health
b
Low 1,240 6.3 (0.4) 4.9 (0.5) 5.6 (0.6) 1.4 (0.4) 0.8 (0.5)
Medium 6,188 45.0 (1.0) 41.9 (1.2) 43.4 (1.3) 3.2 (0.9) 1.7 (1.0)
High 6,270 48.6 (1.0) 53.2 (1.3) 51.0 (1.3) 4.6 (0.9) 2.4 (1.0)
Satisfaction with housing
b
Low 670 2.4 (0.3) 3.0 (0.5) 2.3 (0.5) 0.6 (0.3) 0.5 (0.3)
Medium 4,438 26.2 (1.0) 28.0 (1.2) 26.4 (1.2) 1.7 (0.8) 0.2 (0.9)
High 8,588 71.4 (1.0) 69.0 (1.3) 70.7 (1.3) 2.4 (0.9) 0.7 (0.9)
(continued)
23
Table 4. (continued)
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of Android Smartphone Owners Android Smartphone Coverage Bias
With Standard
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Satisfaction with living standards
b
Low 840 2.7 (0.3) 2.3 (0.4) 2.2 (0.3) 0.4 (0.3) 0.5 (0.3)
Medium 5,900 32.1 (1.1) 34.2 (1.4) 33.4 (1.4) 2.1 (0.9) 1.3 (1.0)
High 6,948 65.2 (1.1) 63.5 (1.4) 64.4 (1.4) 1.8 (1.0) 0.8 (1.0)
Self-efficacy
c
Low 4,582 28.7 (0.9) 27.2 (1.1) 26.9 (1.2) 1.5 (0.8) 1.8 (0.9)
Medium 5,557 46.8 (1.0) 49.7 (1.4) 49.2 (1.4) 2.9 (0.9) 2.3 (1.0)
High 3,358 24.5 (0.8) 23.0 (1.0) 24.0 (1.2) 1.5 (0.7) 0.5 (0.9)
Size of personal network
0–2 2,352 13.0 (0.7) 13.9 (0.9) 13.5 (0.9) 0.9 (0.6) 0.5 (0.6)
3–9 7,369 56.8 (0.9) 58.2 (1.4) 57.0 (1.4) 1.3 (0.9) 0.1 (1.0)
10 or more 3,899 30.0 (1.0) 28.0 (1.3) 29.6 (1.4) 2.2 (0.9) 0.6 (1.0)
Social inclusion
d
Low 933 3.4 (0.4) 2.2 (0.3) 2.2 (0.3) 1.3 (0.3) 1.3 (0.4)
Medium 5,366 34.0 (1.0) 34.1 (1.3) 33.6 (1.3) 0.1 (0.9) 0.3 (0.9)
High 7,305 62.6 (1.0) 64.7 (1.3) 64.2 (1.3) 1.1 (0.9) 1.6 (0.9)
Note: Bold indicates statistically significant bias estimates based on a z-test (p< .05).
a
Deprivation is measured as an index using 23 items available in households weighted based on needfulness ranging from 0 to 8.2. No deprivation (all items
available) ¼0, medium deprivation ¼0.1–0.9, and high deprivation ¼1.0–8.2.
b
Original satisfaction scale ranged from 0 ¼completely unsatisfied to 10 ¼completely satisfied. Low ¼0–3, medium ¼4–7, and high ¼8–10.
c
Based on composite score from 5 items.
d
Original social inclusion scale ranged from 1 ¼excluded to 10 ¼included. Low ¼1–3, medium ¼4–7, and high ¼8–10.
24
Table 5. Weighted Estimates and Standard errors (SE) for Substantive Panel Study Labour Market and Social Security (PASS) Wave 11
Variables Based on iPhone Ownership and Weighted Estimates of iPhone Coverage Bias.
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of iPhone Owners iPhone Coverage Bias
With Standard
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Marital status
Married 6,087 57.4 (1.1) 52.2 (2.3) 58.6 (2.6) 5.2 (2.2) 1.3 (2.6)
Divorced, widowed, separated 2,765 20.7 (1.0) 13.1 (1.6) 17.8 (2.3) 7.7 (1.6) 2.9 (2.4)
Never married 4,700 21.9 (0.7) 34.8 (1.6) 23.6 (1.6) 12.9 (1.8) 1.7 (1.5)
Household (HH) size
1 3,462 23.2 (0.9) 19.0 (1.9) 18.5 (2.0) 4.2 (1.8) 4.7 (2.2)
2 4,661 38.3 (1.1) 32.6 (2.4) 40.4 (2.9) 5.7 (2.3) 2.1 (2.8)
3 2,367 17.4 (0.9) 20.0 (2.0) 18.3 (2.9) 2.6 (2.3) 0.9 (1.8)
4 1,883 14.8 (0.9) 22.4 (2.1) 17.5 (1.7) 7.6 (1.9) 2.7 (1.6)
5 or more 1,330 6.3 (0.6) 6.0 (1.1) 5.3 (1.0) 0.3 (1.1) 1.0 (1.0)
Own children in HH
Yes 4,849 32.2 (1.0) 35.9 (2.4) 33.7 (2.4) 3.7 (2.2) 1.5 (2.2)
HH income
Less than 1,000 euros 4,657 15.9 (0.8) 11.8 (1.3) 11.4 (1.4) 4.2 (1.4) 4.6 (1.6)
1,000–1,999 euros 6,027 49.6 (1.3) 40.3 (2.6) 42.9 (2.8) 9.3 (2.5) 6.7 (2.7)
2,000 euros or more 2,877 34.4 (1.2) 48.0 (2.6) 45.7 (2.8) 13.5 (2.5) 11.3 (2.7)
Employment status
Employed 6,030 51.7 (0.9) 63.3 (2.1) 59.6 (2.7) 11.7 (2.0) 8.0 (2.5)
Unemployed 2,608 4.9 (0.3) 4.1 (0.8) 4.0 (0.8) 0.8 (0.8) 0.9 (0.7)
Inactive 4,979 43.3 (0.9) 32.5 (2.0) 36.3 (2.6) 10.9 (1.9) 7.0 (2.5)
(continued)
25
Table 5. (continued)
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of iPhone Owners iPhone Coverage Bias
With Standard
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Working hours
Under 35 hours 1,808 27.8 (1.0) 27.1 (2.5) 28.9 (2.9) 0.7 (2.3) 1.1 (2.6)
35–40 hours 2,017 34.8 (1.3) 29.7 (2.7) 30.2 (2.7) 5.1 (2.5) 4.6 (2.5)
More than 40 hours 1,906 37.3 (1.3) 43.1 (2.9) 40.9 (3.1) 5.8 (2.6) 3.6 (2.7)
Welfare receipt
Yes 3,367 5.6 (0.4) 4.4 (0.9) 4.5 (0.9) 1.1 (0.9) 1.1 (0.8)
Deprivation
a
No deprivation 5,772 64.8 (1.2) 68.2 (2.4) 68.1 (2.5) 3.4 (2.4) 3.4 (2.5)
Medium deprivation 4,240 24.8 (1.0) 23.8 (2.3) 23.8 (2.4) 1.1 (2.2) 1.0 (2.4)
High deprivation 3,691 10.4 (0.7) 8.1 (1.2) 8.0 (1.2) 2.3 (1.2) 2.4 (1.2)
Satisfaction with life in general
b
Low 670 2.3 (0.3) 1.3 (0.4) 1.3 (0.4) 0.9 (0.5) 0.9 (0.6)
Medium 6,154 38.7 (1.1) 33.2 (2.4) 31.7 (2.3) 5.5 (2.2) 7.0 (2.4)
High 6,860 59.1 (1.2) 65.5 (2.3) 67.0 (2.4) 6.4 (2.2) 7.9 (2.4)
Satisfaction with health
b
Low 1,240 6.3 (0.4) 3.3 (0.7) 3.6 (0.7) 3.1 (0.7) 2.7 (1.0)
Medium 6,188 45.0 (1.0) 37.5 (2.0) 41.9 (2.4) 7.5 (2.0) 3.1 (2.4)
High 6,270 48.6 (1.0) 59.2 (2.0) 54.5 (2.3) 10.6 (2.0) 5.9 (2.3)
Satisfaction with housing
b
Low 670 2.4 (0.3) 1.9 (0.6) 2.0 (0.6) 0.5 (0.6) 0.4 (0.6)
Medium 4,438 26.2 (1.0) 23.7 (2.1) 22.4 (2.1) 2.6 (1.9) 3.9 (2.0)
High 8,588 71.4 (1.0) 74.5 (2.1) 75.7 (2.2) 3.1 (2.0) 4.3 (2.1)
(continued)
26
Table 5. (continued)
Variable
(1)
Unweighted (n)
(2)
Full Sample
(3) (4) (5) (6)
Covered Sample of iPhone Owners iPhone Coverage Bias
With Standard
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
With Standard
PASS Weights
With Adjusted
PASS Weights
Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)
Satisfaction with living standards
b
Low 840 2.7 (0.3) 1.1 (0.4) 1.0 (0.3) 1.6 (0.4) 1.7 (0.4)
Medium 5,900 32.1 (1.1) 25.8 (2.1) 26.0 (2.3) 6.3 (2.0) 6.1 (2.2)
High 6,948 65.2 (1.1) 73.1 (2.1) 73.0 (2.3) 7.9 (2.1) 7.8 (2.3)
Self-efficacy
c
Low 4,582 28.7 (0.9) 26.3 (2.0) 25.8 (2.2) 2.4 (1.9) 2.9 (2.1)
Medium 5,557 46.8 (1.0) 48.9 (2.5) 47.1 (2.6) 2.1 (2.3) 0.3 (2.5)
High 3,358 24.5 (0.8) 24.7 (1.9) 27.1 (2.2) 0.3 (1.8) 2.7 (2.2)
Size of personal network
0–2 2,352 13.0 (0.7) 8.3 (1.2) 8.8 (1.5) 4.7 (1.2) 4.2 (1.5)
3–9 7,369 56.8 (0.9) 60.1 (2.3) 56.9 (2.5) 4.0 (2.2) 0.0 (2.5)
10 or more 3,899 30.0 (1.0) 30.9 (2.3) 34.3 (2.5) 0.7 (2.1) 4.2 (2.4)
Social inclusion
d
Low 933 3.4 (0.4) 1.8 (0.5) 1.8 (0.5) 1.6 (0.6) 1.6 (0.8)
Medium 5,366 34.0 (1.0) 27.8 (2.3) 26.9 (2.5) 6.1 (2.2) 7.0 (2.4)
High 7,305 62.6 (1.0) 70.4 (2.3) 71.2 (2.5) 7.8 (2.2) 8.6 (2.4)
Note: Bold indicates statistically significant bias estimates based on a z-test (p< .05).
a
Deprivation is measured as an index using 23 items available in households weighted based on needfulness ranging from 0 to 8.2. No deprivation (all items
available) ¼0, medium deprivation ¼0.1–0.9, and high deprivation ¼1.0–8.2.
b
Original satisfaction scale ranged from 0 ¼completely unsatisfied to 10 ¼completely satisfied. Low ¼0–3, medium ¼4–7, and high ¼8–10.
c
Based on composite score from 5 items.
d
Original social inclusion scale ranged from 1 ¼excluded to 10 ¼included. Low ¼1–3, medium ¼4–7, and high ¼8–10.
27
owners, we overestimate the proportion of people who report having a per-
sonal network of between three and nine people (þ1.6 p.p.) and underesti-
mate the proportion of people who report having a personal network size of
10 or more people (1.1 p.p.). We find a bias toward German residents
reporting high social inclusion (þ2.4 p.p.) compared to people reporting
medium (1.2 p.p.) and low social inclusion (1.2 p.p.).
Regarding the bias produced by using the sample of respondents who
reported owning an Android smartphone, we find comparable effects with
slightly higher biases than those for the smartphone sample (see Column 5
in Table 4). The proportions of German residents aged 15 and older who
reported never been married (+5.7 p.p.), were living in households with
more than two people, were living with their own children (+8.9 p.p.),
were employed (+10.4 p.p.), had high satisfaction with their health (+4.6
p.p.), had low (+0.6 p.p.) or medium (+1.7 p.p.) satisfaction with their
housing, and had medium self-efficacy (+2.9 p.p.) are overestimated from
the covered sample of Android smartphone owners. We find a negative
bias estimate for people who reported that they were divorced, widowed,
or separated (-6.9 p.p.); were living in two- (-5.8 p.p.) or one-person
households (-5.2 p.p.); were inactive in the labor market (-11.4 p.p.); had
low general life satisfaction (-0.8 p.p.); had low (-1.4 p.p.) or medium
health satisfaction (-3.2 p.p.); had high satisfaction with their housing (-2.4
p.p.); had a social network of more than 10 people (-2.2 p.p.); and had low
social inclusion (-1.3 p.p.). In addition, using the sample of covered
Android owners, we overestimate the percentage of people reporting 35–
40 working hours per week by two points, and we underestimate the
percentage of people working under 35 hours per week (1.7 p.p.). We
also see a bias in reported satisfaction with living standards; the proportion
of people reporting medium satisfaction is overestimated by two points,
while the number of people reporting high satisfaction is underestimated
(1.8 p.p.).
Finally, Table 5 (Column 5) shows that we find even larger biases when
we use the sample of iPhone owners to estimate the substantive PASS mea-
sures. For example, the percentages of never married German residents,
those living in households in the highest income bracket, and those being
employed are all overestimated by more than 13 points. In addition, we find
that the percentage of people working more than 40 hours per week is over-
estimated by almost six points. We also find relatively large biases in the
satisfaction measures, with an overestimation of people who reported having
high satisfaction with life in general (þ6.4 p.p.), health (þ10.6 p.p.), and
living standards (þ7.9 p.p.). Finally, the percentage of people who reported
28 Sociological Methods & Research XX(X)
having high social inclusion is overestimated by almost eight points com-
pared to the percentages of people who reported having low (1.6 p.p.) and
medium social inclusion (6.1 p.p.).
Can Coverage Bias Be Reduced by Implementing Specific Weights
That Correct for Known Differences in Device Ownership?
Columns 4 in Tables 3–5 present the estimates and standard errors from the
reduced samples of smartphone, Android smartphone, and iPhone owners,
which were calculated with the newly adjusted PASS weights accounting for
sociodemographic differences in device ownership. Column 6 provides the
bias estimates and corresponding standard errors that resulted when the full
PASS sample estimates were compared with the covered sample using the
adjusted PASS weights.
Column 6 in Table 3 shows that the newly adjusted PASS weights using
sociodemographic variables can reduce but not eliminate coverage bias that
stems from smartphone ownership. Differences in estimates of marital status,
household size, prevalence of individuals who live with their own children in
the household, satisfaction with health and housing, and personal network
size are reduced. In contrast, biases in household income, general life satis-
faction and living standards, and self-efficacy due to smartphone ownership
are unaffected by adjusting the PASS weights. Nevertheless, none of these
biases are larger than 3.5 percentage points after the adjusted PASS weights
are applied.
We see a similar pattern for Android smartphone coverage bias (Column 6
in Table 4); most of the biases are reduced when we apply the new weights
adjusting for sociodemographic differences between Android smartphone
owners and the general population. Nevertheless, the percentage of
employed people is still overestimated by more than six percentage points,
and the percentage of inactive German residents is still underestimated by
almost seven points. Biases in attitudinal variables and social embeddedness
are mostly under two percentage points.
Finally, applying the iPhone-specific adjusted weights reduces most of the
bias in other sociodemographic variables, but we still find large biases in
many of the other measures. For example, the percentage of people from
high-income households is overestimated by more than 11 points, and the
percentage of employed German residents is overestimated by 8 points.
Similarly, we still see a large bias in the percentage of people who reported
having high general life satisfaction (þ7.9 p.p.), high satisfaction with their
living standards (þ7.8 p.p.), and high social inclusion (þ8.6 p.p.).
Keusch et al. 29
When comparing the standard errors between the estimates using the
standard PASS weights (Column 3 in Tables 3 through 5) and the estimates
using the adjusted PASS weights (Column 4), the size of the standard errors
is relatively unaffected by the additional weighting step. There seems to be
no trade-off between a reduction in bias and an increase in the standard
errors, suggesting that a reduction in bias will also lead to a reduction in
mean squared error (not presented here). It thus seems advisable to apply the
additional weighting step for sociodemographic differences when analyzing
a sample of smartphone owners.
Conclusion
Using data from the PASS, an annual probability-based mixed-mode survey
on labor market and poverty that targets the residential population of Ger-
many aged 15 and older, we estimate that in 2017, 76 percent of German
residents own a smartphone, 49 percent own an Android smartphone, and 17
percent own an iPhone. In accordance with the results of previous research
on other European countries (Baier et al. 2018; Fuchs and Busse 2009;
Metzler and Fuchs 2014) and the United States (Couper et al. 2018; Pew
Research 2017), we find that smartphone owners are overrepresented among
young and highly educated people and those who live in large communities.
We also find that smartphone penetration is higher among non-German
citizens than among German citizens. In addition, we find that smartphone
ownership in Germany correlates with a number of substantive variables on
the labor market and poverty, which are routinely collected as part of PASS.
Our findings confirm the notion that the digital divide between owners and
nonowners of smartphones expands beyond mere sociodemographic differ-
ences and affects measures of behaviors and attitudes.
Using weighting techniques based on the sociodemographic information
on age, sex, educational attainment, nationality, region, and community size
that is available for both owners and nonowners of smartphones in Germany
does reduce some of the differences but does not eliminate them. The biases
created by differences in estimates from the full sample and the sample of
smartphone owners are statistically significant for many of the PASS vari-
ables. Some of the biases on attitudinal variables, such as satisfaction with
different aspects of one’s life, social inclusion, and self-efficacy, are sub-
stantially small, with most of the bias estimates not exceeding two percent-
age points. Since some of the coverage bias is clearly driven by very low
smartphone penetration in the older age groups, we see an improvement in
bias for measures that highly correlate with age (e.g., satisfaction with
30 Sociological Methods & Research XX(X)
health) when applying additional weights that account for these age differ-
ences. This finding confirms the results reported by Couper et al. (2018) on
smartphone coverage bias on measures of fertility and sexual behavior from
the NSFG in individuals aged 15–44 in the United States. Taken together,
these findings provide additional evidence that smartphone penetration
seems to have reached a level in many Western countries that is comparable
to the level of nonmobile Internet coverage. Thus, conducting mobile web
surveys with smartphones among people under the age of 65 seems to pro-
duce an amount of coverage bias that is not very different from that produced
by traditional web surveys limited to landline Internet.
However, other substantive sociodemographic measures, such as house-
hold size, prevalence of individuals living with their own children in the
household, household income, and employment status, show larger sub-
stantial biases, even after weights that specifically account for sociodemo-
graphic differences between owners and nonowners of the device are
applied. These biases might be driven by other mechanisms, such as dif-
ferences in income and other wealth-related variables between people who
own and do not own a smartphone.
While most mobile browser-based web surveys currently are OS agnostic,
that is, there are no major restrictions by the specific OS that is installed on a
smartphone regarding the capability of displaying and responding to a web
survey in a mobile browser, other forms of data collection using smartphones
remain susceptible to the idiosyncrasies of different types of smartphones.
For example, Android as an open-source platform is currently less restrictive
in how and what data from different sensors and other apps can be accessed
by and collected through a research app than the iOS ecosystem on Apple
iPhones. If researchers want to leverage the full breadth of passive measure-
ments in one study (e.g., continuous measurements of geolocation and move-
ments, timestamps of app and browser use, and call and text message logs) as
described by Kreuter et al. (2018), the researches would need to rely on
participants with Android smartphones. Our study is the first to consider the
bias that stems from coverage of smartphones with different OSs. For the
substantive variables of PASS, we find that the bias produced by Android
smartphone coverage is generally not much higher than the coverage bias for
all smartphones in Germany, which is encouraging news for researchers who
are eager to use smartphone technology for passive measurements. However,
iPhone owners constitute a much smaller and more unique subpopulation in
Germany, both in terms of sociodemographic characteristics and in terms of
attitudes and behaviors, which confirms the results of previous research (e.g.,
Go¨tz et al. 2017; Pryss et al. 2018; Shaw et al. 2016; Ubhi et al. 2017). In our
Keusch et al. 31
study, we find relatively low iPhone coverage and large biases in many of the
substantive PASS measures included in our analysis when we restrict the
sample to iPhone owners. This bias is substantial even when iPhone-specific
weights that should correct for sociodemographic differences between
iPhone owners and the general population are applied. Thus, we caution
researchers to limit their data collection to iPhone owners when possible
since this strategy can systematically exclude specific subpopulations (e.g.,
individuals who are older, are not single, have a low income, work a small
number of hours, have low satisfaction with various aspects of life, feel
deprived, and feel less socially embedded).
Of course, the way we and others operationalize coverage assumes
that ownership of a device equals uniform use of the technology across
all owners. From research on how people engage with the Internet and
other IT technology, we know that access to a technology does not
necessarily mean that every individual is able and willing to use the
technology to its full potential. Hargittai (2002) uses the term “second
digital divide” to describe this phenomenon. Recent studies have found
that smartphone skills and the use of smartphones for different activities
correlate with the reported willingness to participate in smartphone data
collection tasks that are more complex than mobile web surveys (Couper
et al. 2017; Keusch, Struminskaya, et al. 2019; Wenz, Ja¨ckle, and Couper
2017). In addition, privacy and security concerns predict how willing
smartphone users are to share passive data with researchers (Ja¨ckle
et al. 2019; Keusch, Struminskaya, et al. 2019; Revilla, Couper, and
Ochoa 2019; Wenz et al. 2017). In the future, we hope to see more
research on selective participation in studies involving smartphone data
collection and the biases associated with nonparticipation.
It is important to keep in mind that accessibility and willingness varies
across study topics and institutions conducting the research. Thus, our study
also needs to be interpreted in its context. We researched biases with data
from a study on labor market activities and poverty in Germany. Coverage
biases might look different when different outcomes are used as the variables
of interest. Furthermore, smartphone contracts and devices vary in availabil-
ity across countries. Before one generalizes our results from Germany to a
different country, we recommend that they consider the possible relation-
ships between smartphone and OS usage and their correlations with the
substantive variables examined here. Our findings indicate that age and
income as well as attitudinal measures on life satisfaction are strong corre-
lates with smartphone ownership and thus need to be considered when smart-
phones are used as the main mode for collecting information that might be
32 Sociological Methods & Research XX(X)
associated with these variables. We encourage others to also add to the small
body of literature on coverage errors and coverage bias, so we can collec-
tively learn about the error properties of data collected with smartphones and
other mobile devices.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research,
authorship, and/or publication of this article.
Funding
The author(s) declared the following potential conflicts of interest with respect to the
research, authorship, and/or publication of this article: Georg-Christoph Haas’ time
was supported by the German Research Foundation (DFG) through the Collaborative
Research Center SFB 884 “Political Economy of Reforms” (Project A9) [139943784
to Markus Fro¨lich, Florian Keusch, and Frauke Kreuter].
ORCID iD
Florian Keusch https://orcid.org/0000-0003-1002-4092
Notes
1. In Waves 1–4 of PASS, CATI was used as the main mode, and CAPI was used as
the follow-up mode. Since Wave 5, this order has been reversed.
2. All questionnaires and a detailed user guide (Bethmann, Fuchs, and Wurdack
2017) can be found on the website of the Research Data Center of the Federal
Employment Agency at Institute for Employment Research (http://fdz.iab.de/en/
FDZ_Individual_Data/PASS.aspx).
3. Deprivation is measured as an index using 23 goods (e.g., car, washing machine) or
activities (e.g., one-week vacation, inviting friends over for dinner) that the house-
holds can or cannotafford. The goods and activities that a householdcannot afford are
weightedbased on their perceived importance for a good life. The indexranges from 0
to 8.2. We used the following three categories for our analysis: no deprivation (all
items available) ¼0, medium deprivation ¼0.1–0.9, and high deprivation ¼1.0–8.2.
4. The satisfaction items were measured on an 11-point end-labeled scale ranging
from 0 ¼completely unsatisfied to 10 ¼completely satisfied. We use the follow-
ing categories for our analysis: low ¼0–3, medium ¼4–7, and high ¼8–10.
5. Self-efficacy was measured based on 5 items. Principal component factor analysis
identified that all 5 items loaded on one common factor. We used regression to
predict factor scores, centered means, and built three equally large groups from the
metric factor.
Keusch et al. 33
6. Social inclusion was measured on a 10-point end-labeled scale ranging from 1 ¼
excluded to 10 ¼included. We use the following categories for our analysis:
low ¼1–3, medium ¼4–7, high ¼8–10.
References
Antoun, Christopher, Frederick G. Conrad, Mick P. Couper, and Brady T. West.
2018. “Simultaneous Estimation of Multiple Sources of Error in a Smartphone-
based Survey.” Journal of Survey Statistics and Methodology 7:93-117.
Baier, Tobias, Anke Metzler, and Marek Fuchs. 2018. “Coverage Error in Mobile
Web Surveys across European Countries 2014-2017.” Presented at the 73rd
Annual Conference of the America Association for Public Opinion Research, May
16–19, 2018, Denver, CO.
Berg, Marco, Ralph Cramer, Christian Dickmann, Reiner Gilberg, Birgit Jesske,
Martin Kleudgen, Jonas Beste, Sandra Dummert, Corinna Frodermann, Stefan
Schwarz, and Mark Trappmann. 2018. Codebook and Documentation of the Panel
Study “Labour Market and Social Security” (PASS). Datenreport Wave 11. (FDZ-
Datenreport, XX/2018 (en)), Nuremberg, Germany.
Beste, Jonas, Johannes Eggs, Stefanie Gundert, and Claudia Wenzig. 2013.
“Instruments and Interview Programme.” Pp. 23-26 in User Guide “Panel Study
Labour Market and Social Security” (PASS). Wave 6. (FDZ-Datenreport, 07/2013
(en)), edited by A. Bethmann, B. Fuchs, and A. Wurdack. Nuremberg, Germany:
IAB.
Bethmann, Arne, Benjamin Fuchs, and Anja Wurdack. 2017. User Guide “Panel
Study Labour Market and Social Security” (PASS). Wave 6. (FDZ-Datenreport,
07/2013 (en)). Nuremberg, Germany: IAB.
Church, Karen, Denzil Ferreira, Nikola Banovic, and Kent Lyons. 2015.
“Understanding the Challenges of Mobile Phone Use Data.” Proceedings of the
MobileHCI ‘15, 504-14. doi: 10.1145/2785830.2785891.
Couper, Mick P., Christoper Antoun, and Aigul Mavletova. 2017. “Mobile Web
Surveys: A Total Survey Error Perspective.” Pp. 133-54 in Total Survey Error
in Practice, edited by Biemer, Paul P., Edith D. de Leeuw, Stephanie Eckman,
Brad Edwards, Frauke Kreuter, Lars E. Lyberg, N. Clyde Tucker, and Brady T.
West. Hoboken, NJ: Wiley.
Couper, Mick P., Garret Gremel, William Axinn, Heidi Guyer, James Wagner, and
Brady T. West. 2018. “New Options for National Population Surveys: The Impli-
cations of Internet and Smartphone Coverage.” Social Science Research 73:
221-35.
Dissanayake, Ravindra, and Thustan Amarasuriya. 2015. “Role of Brand Identity in
Developing Global Brands: A Literature Based Review on Case Comparison
34 Sociological Methods & Research XX(X)
between Apple iPhone vs. Samsung Smartphone Brands.” Research Journal of
Business and Management 2:430-40.
Eckman, Stephanie and Frauke Kreuter. 2017. “The Undercoverage-nonresponse
Tradeoff.” Pp. 97-113 in Total Survey Error in Practice, edited by Biemer, Paul
P., Edith D. de Leeuw, Stephanie Eckman, Brad Edwards, Frauke Kreuter, Lars E.
Lyberg, N. Clyde Tucker, and Brady T. West. Hoboken, NJ: Wiley.
Eurostat. 2018. Digital Economy and Society: ICT usage in Households and by
Individuals. Retrieved May 30, 2018 (http://ec.europa.eu/eurostat/web/digital-
economy-and-society/data/database).
Fortney, John C., James F. Burgess, Hayden B. Bosworth, Brenda M. Booth, and
Peter J. Kaboli. 2011. “A Re-conceptualization of Access for 21st Century Health-
care.” Journal of General Internal Medicine 26:S639-47.
Fuchs, Marek and Britta Busse. 2009. “The Coverage Bias of Mobile Web Sur-
veys across European Countries.” International Journal of Internet Science 4:
21-33.
Go¨tz, Friedrich M., Stefan Stieger, and Ulf-Dietrich Reips. 2017. “Users of the Main
Smartphone Operating Systems (iOS, Android) Differ Only Little in Personality.”
PLoS One 12(5):e0176921. doi: 10.1371/journal.pone.0176921.
Groves, Robert, Floyd Fowler, Mick P. Couper, James Lepkowski, Eleanor Singer,
and Roger Tourangeau. 2009. Survey Methodology, 2nd ed. New York: Wiley.
Hargittai, Esther. 2002. “Second-level Digital Divide: Differences in People’s Online
Skills.” First Monday 7(4). Retrieved March 30, 2018 (http://firstmonday.org/
issues/issue7_4/hargittai/index.html).
Ja¨ckle, Annette, John Burton, Mick P. Couper, and Carli Lessof. 2019. “Participation
in a Mobile App Survey to Collect Expenditure Data as Part of a Large-scale
Probability Household Panel: Coverage and Participation Rates and Biases.” Sur-
vey Research Methods 13:23-44.
Jesske, Birgit and Sabine Schulz. 2018. Methodenbericht: Panel Arbeitsmarkt und
Soziale Sicherung. PASS-Erhebungswelle 11 – 2017. (Haupterhebung). (FDZ-
Methodenreport XX/2018). Nuremberg, Germany: IAB.
Kantar. 2017. Sales Market Share of Leading Smartphone Operating Systems in
Germany from August to October 2016 and 2017. Retrieved May 30, 2018
(https://www.statista.com/statistics/461997/smartphone-os-market-shares-of-
sales-in-germany/).
Keusch, Florian, Mariel M. Leonard, Christoph Sajons, and Susan Steiner. 2019.
“Using Smartphone Technology for Research on Refugees—Evidence from Ger-
many.” Sociological Methods & Research, May 30. https://doi.org/10.1177/
0049124119852377.
Keusch et al. 35
Keusch, Florian, Bella Struminskaya, Christopher Antoun, Mick P. Couper, and
Frauke Kreuter. 2019. “Willingness to Participate in Passive Mobile Data Collec-
tion.” Public Opinion Quarterly 83: 210-35.
Keusch, Florian and Ting Yan. 2017. “ Web vs. Mobile Web—An Experimental
Study of Device Effects and Self-selection Effects.” Social Science Computer
Review 35:147-66.
Kreuter, Frauke, Haas Georg-Christoph, Keusch Florian, Ba¨ hr Sebastian, and Trapp-
mann Mark. 2018. “Collecting survey and smartphone sensor data with an app:
Opportunities and challenges around privacy and informed consent.” Social Sci-
ence Computer Review, December 18, 2018. https://doi.org/10.1177/08944393
18816389
Lane, Wilburn and Chris Manner. 2011. “The Impact of Personality Traits on Smart-
phone Ownership and Use.” International Journal of Business and Social Science
2(17):22-28.
Metzler, Anke and Marek Fuchs. 2014. “Coverage Error in Mobile Web Surveys
across European Countries.” Presented at the 7th Internet Survey Methodology
Workshop, December 1–3, 2014, Bozen-Bolzano, Italy.
Park, Eun-A. and Sangwon Lee. 2015. “Multidimensionality: Redefining the Digital
Divide in the Smartphone Era.” Info 17(2):80-96.
Pew Research Center. 2017. Mobile Fact Sheet. Washington, DC: Pew Research
Center. Retrieved May 30, 2018 (http://www.pewinternet.org/fact-sheet/mobile/).
Pryss, Ru
¨diger, Manfred Reichert, Winfried Schlee, Myra Spiliopoulou, Berthold
Langguth, and Thomas Probst. 2018. “Differences between Android and iOS
Users of the TrackYourTinnitus Mobile Crowdsensing mHealth Platform.” Pro-
ceedings of the 2018 IEEE 31st International Symposium on Computer-based
Medical Systems (CBMS), 411-16. doi: 10.1109/CBMS.2018.00078
Revilla, Melanie, Mick P. Couper, and Carlos Ochoa. 2019. “Willingness of Online
Panelists to Perform Additional Tasks.” Methods, Data, Analyses 13:223-52.
Rudolph, Helmut and Mark Trappmann. 2007. “Design und Stichprobe des Panels
Arbeitsmarkt und Soziale Sicherung (PASS).” Pp. 60-101 in Neue Daten fu
¨r die
Sozialstaatsforschung. Zur Konzeption der IAB-Panelerhebung “Arbeitsmarkt
und Soziale Sicherung”. (IAB-Forschungsbericht, 12/2007),editedbyM.
Promberger. Nuremberg, Germany: IAB.
Scherpenzeel, Annette. 2017. “Mixing Online Panel Data Collection with Innovative
Methods.” Pp. 27-49 in Methodische Probleme von Mixed-Mode-Ansa¨ tzen in der
Umfrageforschung, edited by S. Eifler and F. Faulbaum. Wiesbaden, Germany:
Springer.
Schmall, Tyler.2018. “iPhone and Android Users Are Completely Different People.”
October 25. Retrieved (https://nypost.com/2018/10/25/iphone-and-android-users-
are-completely-different-people/)
36 Sociological Methods & Research XX(X)
Shaw, Heather, David A. Ellis, Libby-Rae Kendrick, Fenja Ziegler, and Richard
Wiseman. 2016. “Predicting Smartphone Operating System from Personality and
Individual Differences.” Cyberpsychology, Behavior, and Social Networking, 19:
727-32.
Sugie, Naomi F. 2018. “Utilizing Smartphones to Study Disadvantaged and Hard-to-
reach Groups.” Sociological Methods & Research 47:458-91.
Trappmann, Mark. 2013. “Weights.” Pp. 81-99 in User Guide “Panel Study
Labour Market and Social Security” (PASS). Wave 6. (FDZ-Datenreport,
07/2013 (en)), edited by A. Bethmann, B. Fuchs, and A. Wurdacks. Nurem-
berg, Germany: IAB.
Trappmann, Mark, Sebastian Ba¨ hr, Jonas Beste, Andreas Eberl, Corinna Frodermann,
Stefanie Gundert, Stefan Schwarz, Nils Teichler, Stefanie Unger, and Claudia
Wenzig. (2019): Data Resource Profile: Panel Study Labour Market and Social
Security (PASS). International Journal of Epidemiology. Retrieved (https://doi.
org/10.1093/ije/dyz041).
Trappmann, Mark, Gerrit Mu
¨ller, and Arne Bethmann. 2013. “Design of the Study.”
Pp. 13-22 in User Guide “Panel Study Labour Market and Social Security”
(PASS). Wave 6. (FDZ-Datenreport, 07/2013 (en)), edited by A. Bethmann, B.
Fuchs, and A. Wurdack. Nuremberg, Germany: IAB.
Ubhi, Harveen Kaur, Daniel Kotz, Susan Michie, Onno C.P. van Schayck, and Robert
West. 2017. “A Comparison of the Characteristics of iOS and Android Users of a
Smoking Cessation App.” Translational Behavioral Medicine 7:166-71.
Valliant, Richard, Jill A. Dever, and Frauke Kreuter. 2013. Practical Tools for
Designing and Weighting Survey Samples. New York: Springer.
Wenz, Alexander, Annette Ja¨ckle, and Mick P. Couper. 2017. “Willingness to Use
Mobile Technologies for Data Collection in a Probability Household Panel.”
Survey Research Methods 13:1-22.
Yan, Ting, Janice Machado, Andrew Heller, Erika Bonilla, Aaron Maitland, and John
Kirlin. 2017. “The Feasibility of Using Smartphones to Record Food Purchase and
Acquisition.” Presented at the AAPOR 72th Annual Conference, May 18–21,
2017, New Orleans, Louisiana.
Yarrow, Jay. 2013. “Here’s A Problem with the Theory That Android Is Taking
Over the World.” November 30. Retrieved (https://www.businessinsider.com/
ios-android-shopping-traffic-2013-11?IR¼T).
Author Biographies
Florian Keusch is Professor (interim) for Statistics and Methodology at the Univer-
sity of Mannheim, Germany and Adjunct Assistant Professor at the Joint Program in
Survey Methodology (JPSM) University of Maryland, USA. He received a PhD in
Keusch et al. 37
social and economic sciences (Dr.rer.soc.oec.) and an MSc in business (Mag.rer.so-
c.oec.) from WU, Vienna University of Economics and Business. His research
focuses on nonresponse and measurement error in Web and mobile Web surveys,
passive mobile data collection, and visual design effects in questionnaires.
Sebastian Ba
¨hr is a senior researcher at the research department “Panel Study
Labour Market and Social Security” (PASS) at the Institute for Employment
Research (IAB) in Nuremberg, Germany. He holds a B.A. in international business
from the University of Hull, United Kingdom, as well as a master’s degree in social
economics and a PhD (Dr.rer.pol) from the School of Business and Economics of the
Friedrich-Alexander University Erlangen-Nu
¨rnberg (FAU), Germany. His primary
research interests are social inequality, labor markets, social networks, and unem-
ployment. He studies social networks and their implications on labor market
behavior.
Georg-Christoph Haas is a research associate at the Statistical Methods Centre of
the Institute for Employment Research (IAB) where he contributes in the project
“Quality in Establishment Surveys” (QuEst) and “Mobile Device Measures”
(MoDeM) and a researcher at the Collaborative Research Center 884 “Political Econ-
omy of Reforms” (SFB 884), University of Mannheim, Germany where he contri-
butes to the project A9 “Survey mode, survey technology and technology innovations
in data collection”. He studied sociology (Diplom) at TU Dresden, Germany.
Frauke Kreuter is Professor in the Joint Program in Survey Methodology at the
University of Maryland, USA, Professor of Methods and Statistics at the University of
Mannheim, Germany, and head of the statistical methods group at the German Insti-
tute for Employment Research (IAB), Nuremberg. She is co-founder of the Coleridge
Initiative and the International Program in Survey and Data Science (IPSDS). Her
recent textbooks include Big Data and Social Science: A Practical Guide to Methods
and Tools, and Practical Tools for Designing and Weighting Survey Samples.
Mark Trappmann is in charge of the Panel Study “The Labour Market and Social
Security” (PASS) at the Institute for Employment Research (IAB) at Nuremberg,
Germany and a Professor for Sociology, especially Survey Methodology, at Otto-
Friedrich-University Bamberg, Germany. He completed his first state teacher training
examination (1. Staatsexamen fu
¨r das Lehramt) in social sciences and mathematics
after studying at the Gerhard-Mercator University in Duisburg, Germany and the
Rijks University in Groningen, The Netherlands, and he received his doctorate
(Dr.Phil.) from the University of Essen, Germany.
38 Sociological Methods & Research XX(X)
... If the willing and unwilling sample members differ systematically on an attribute measured by the sensor, the results will be biased. Several studies have investigated willingness to collect sensor data (Pinter 2015;Scherpenzeel 2017;Keusch et al. 2019;Revilla, Couper, and Ochoa 2019;Wenz, Jäckle, and Couper 2019;Kreuter et al. 2020). However, they too used data from panel studies, possibly inflating participants' motivation and trust in the research (Hillygus, Jackson, and Young 2014;Matthijsse, De Leeuw, and Hox 2015). ...
... However, they too used data from panel studies, possibly inflating participants' motivation and trust in the research (Hillygus, Jackson, and Young 2014;Matthijsse, De Leeuw, and Hox 2015). Moreover, previous studies have focused on hypothetical willingness to share sensor data (Keusch et al. 2019;Wenz et al. 2019). How hypothetical willingness relates to actual participation/sharing is unclear. ...
... To account for this variability in willingness, several studies have focused on the mechanisms underlying willingness to participate in smartphone-sensor data collection (e.g., Keusch et al. 2019;Wenz et al. 2019;Struminskaya et al. 2020). Generally, they point to study, task, and respondent characteristics that might influence stated willingness to share sensor data. ...
Article
Full-text available
Smartphone sensors allow measurement of phenomena that are difficult or impossible to capture via self-report (e.g., geographical movement, physical activity). Sensors can reduce respondent burden by eliminating survey questions and improve measurement accuracy by replacing/augmenting self-reports. However, if respondents who are not willing to collect sensor data differ on critical attributes from those who are, the results can be biased. Research on the mechanisms of willingness to collect sensor data mostly comes from (nonprobability) online panels and is hypothetical (i.e., asks participants about the likelihood of participation in a sensor-based study). In a cross-sectional general population randomized experiment, we investigate how BELLA STRUMINSKAYA is an assistant professor in the features of the request and respondent characteristics influence willingness to share (WTS) and actually sharing smartphone-sensor data. We manipulate the request to either mention or not mention (1) how participation will benefit the participant, (2) participants' autonomy over data collection, and (3) that data will be kept confidential. We assess nonparticipation bias using the administrative records. WTS and actually sharing varies by sensor task, participants' autonomy over data sharing, their smartphone skills, level of privacy concerns, and attitudes toward surveys. Fewer people agree to share photos and a video than geolocation, but all who agreed to share photos or a video actually did. Some nonresponse and nonparticipation biases are substantial and make each other worse, but others jointly reduce the overall bias. Our findings suggest that sensor-data-sharing decisions depend on sample members' situation when asked to share and the nature of the sensor task rather than the sensor type.
... Recent data from early 2021 show that 96% of U.S. adults under the age of 30 but only 61% of adults aged 65 and older own a smartphone. Smartphone ownership increases with household income and educational attainment, and U.S. adults living in rural areas are less likely to own a smartphone than adults living in urban or suburban areas (Pew Research Center, 2021), with similar patterns in other Western countries (Eurostat, 2021;Keusch, Bähr, Haas, Kreuter, & Trappmann, 2020). Beyond disparities in digital access, digital divide scholars have increasingly paid attention to inequalities related to digital skills and the usage of digital technologies, which is often referred to as the second-level digital divide (DiMaggio, Hargittai, Celeste, & Shafer, 2004;Hargittai, 2002). ...
... Smartphone owners who belong to one of these usage types tend to be younger and have higher levels of educational attainment, which is in line with previous research in this area (Andone et al., 2016;Fortunati & Taipale, 2014;Serrano-Cinca et al., 2018). Advanced users also have the highest proportion of iPhone owners, which is consistent with recent research showing that iPhone users tend to be younger and have higher levels of educational attainment (Keusch et al., 2020). ...
Preprint
Full-text available
This paper examines inequalities in the usage of smartphone technology based on five samples of smartphone owners collected in Germany and Austria between 2016 and 2020. We identify six distinct types of smartphone users by conducting latent class analyses that classify individuals based on their frequency of smartphone use, self-rated smartphone skills, and activities carried out on their smartphone. The results show that the smartphone usage types differ significantly by sociodemographic and smartphone-related characteristics: The types reflecting more frequent and diverse smartphone use are younger, have higher levels of educational attainment, and are more likely to use an iPhone. Overall, the composition of the latent classes and their characteristics are robust across samples and time.
... Designing, conducting, and evaluating various forms of data collection processes such that the acquired sample resembles the target population of interest is a core task of survey research. Recent work in survey research investigates coverage and representation issues in the context of digital data and data collected via smartphones and sensors and introduces methods for adjusting non-random samples (Baker et al., 2013;Japec et al., 2015;Keusch et al., 2020). This includes, for example, pseudo-weighting approaches that allow to correct for biases due to selective participation by leveraging information from an auxiliary reference sample (Elliott and Valliant, 2017). ...
Article
Full-text available
Academic and public debates are increasingly concerned with the question whether and how algorithmic decision-making (ADM) may reinforce social inequality. Most previous research on this topic originates from computer science. The social sciences, however, have huge potentials to contribute to research on social consequences of ADM. Based on a process model of ADM systems, we demonstrate how social sciences may advance the literature on the impacts of ADM on social inequality by uncovering and mitigating biases in training data, by understanding data processing and analysis, as well as by studying social contexts of algorithms in practice. Furthermore, we show that fairness notions need to be evaluated with respect to specific outcomes of ADM systems and with respect to concrete social contexts. Social sciences may evaluate how individuals handle algorithmic decisions in practice and how single decisions aggregate to macro social outcomes. In this overview, we highlight how social sciences can apply their knowledge on social stratification and on substantive domains of ADM applications to advance the understanding of social impacts of ADM.
... Moreover, operating systems vary in the amount of information that third parties can access, for example, through research apps. The Android smartphone ecosystem, for example, is much more permissive in the collection of browsing data than Apple's iOS system (Keusch et al., 2020). In addition, ownership of smartphone type co-varies with socio-economic status such that differences in the amount of data that can be collected due to smartphone brand will correlate with participant characteristics. ...
Article
Full-text available
Augmenting survey data with digital traces is a promising direction for combining the advantages of active and passive data collection. However, extracting interpretable measurements from digital traces for social science research is challenging. In this study, we demonstrate how to obtain measurements of news media consumption from survey respondents' web browsing data using Bidirectional Encoder Representations from Transformers, a powerful natural language processing algorithm that estimates contextual word embeddings from text data. Our approach is particularly relevant for political scientists and communication researchers studying exposure to online news content but can easily be adapted to projects in other disciplines working with similar data sets.
... There are different reasons why people might not participate in smartphone data collection . First, access to a smartphone is required, thus creating a problem of coverage: while some of the sociodemographic differences between owners and non-owners of smartphones seem to have decreased over time (Baier et Taylor & Silver, 2019), smartphone owners are younger and more affluent than non-owners and also differ on substantive variables Couper et al., 2018;Jäckle et al., 2019;Keusch et al., 2020). Second, even if smartphone penetration were universal, some smartphone users might be more likely than others to participate in research on their mobile device. ...
Article
Full-text available
Research apps allow to administer survey questions and passively collect smartphone data, thus providing rich information on individual and social behaviours. Agreeing to this novel form of data collection requires multiple consent steps, and little is known about the effect of non-participation. We invited 4,293 Android smart-phone owners from the German Panel Study Labour Market and Social Security (PASS) to download the IAB-SMART app. The app collected data over six months through (a) short in-app surveys and (b) five passive mobile data collection functions. The rich information on PASS members from previous survey waves allows us to compare participants and non-participants in the IAB-SMART study at the individual stages of the participation process and across the different types of data collected. We find that 14.5 percent of the invited smartphone users installed the app, between 12.2 and 13.4 percent provided the different types of passively collected data, and 10.8 percent provided all types of data at least once. Likelihood to participate was smaller among women, decreased with age and increased with educational attainment, German citizen-This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
... Finally, our study's scope was limited to users of Android smartphones, excluding those owning iOS devices. However, as previous studies found no meaningful differences in demographic and personality characteristics between Android and iOS users, this bias should not be too severe (Götz et al., 2017;Keusch et al., 2020). Third, our music preference variables depended on the availability of external songlevel information (i.e., Spotify's audio metrics and Genius' lyrics), which possibly resulted in an underrepresentation of uncommon songs and restricted prediction accuracies for participants with an exotic taste in music. ...
Preprint
Full-text available
Psychologists have long theorized that people use music to create auditory environments matching their personality traits. While there is initial evidence relating self-reported musical style preferences to the Big Five dimensions, little is known about day-to-day music listening behavior and the intrinsic attributes of music that give rise to personality patterns. The present study (N = 330) proposes a personality computing approach to fill these gaps with new insights from ecologically valid music listening records from smartphones. We provided a holistic account of individual differences in music listening by integrating everyday preferences for various musical attributes with habitual listening behaviors. More specifically, we quantified music preferences at fine granularity via technical audio features from Spotify and via lyrical attributes obtained through natural language processing. Using machine learning algorithms, these behavioral variables served to predict Big Five personality on domain and facet level. Our out-of-sample prediction accuracies revealed that the Openness dimension was most strongly related to music listening, while several other traits (most notably Conscientiousness facets) also showed moderate effects. Thereby, preferences for audio and lyrics characteristics were distinctly predictive of personality, hinting at the incremental value of both musical components. Furthermore, variable importance metrics displayed generally trait-congruent relationships between personality outcomes and music listening behaviors prompting us to discuss possible mechanisms underlying these interactions. Overall, our study contributes to the development of a detailed cumulative theory on music listening in personality science, which may be extended in numerous ways in future studies leveraging the computational framework proposed here.
... One major advantage of collecting data through designated research apps is that this allows researchers to specifically design all aspects of the data collection process (e.g., field period, participants' characteristics, particular sensors used) with a specific research question in mind. The controlled environment, potentially in conjunction with a probability sample, allows the researcher to not only to assess coverage (Keusch, Bähr, Haas, Kreuter, & Trappmann, 2020a), nonresponse (Keusch, Bähr, Haas, Kreuter, & Trappmann, 2020b), and measurement error but also to address these issues through weighting techniques known from survey research that would allow inference of the results to a larger population. ...
Article
Smartphones have become central to our daily lives and are often present in the same contexts as their users. Researchers take advantage of this phenomenon by using data from smartphone sensors to infer everyday activities, such as mobility, physical activity, and sleep. For example, that a person is sleeping might be inferred from the fact that their phone is idle and that there is no sound and light around the phone. The success of inference from raw smartphone sensor data to activity outcomes depends, among other factors, on how smartphone owners use their device. Not having the smartphone in close proximity throughout the day, turning the device off, or sharing the device with others can constitute barriers that interfere with accurately measuring everyday activity with data from the phone's native sensors. Against this background, we surveyed two independent, large-scale samples of German smartphone owners (n1 = 3956; n2 = 2525) on how they use their smartphones, with a focus on three everyday activities: mobility, physical activity, and sleep. We find that both sociodemographic as well as smartphone-related characteristics are associated with how people use their smartphones, and that this affects the suitability of smartphone data for measuring everyday activities.
Article
With the ubiquity of smartphones, it is possible to collect self-reports as well as to passively measure behaviors and states (e.g., locations, movement, activity, and sleep) with native sensors and the smartphone’s operating system, both on a single device that usually accompanies participants throughout the day. This research synthesis brings structure to a rapidly expanding body of literature on the combined collection of self-reports and passive measurement using smartphones, pointing out how and why researchers have combined these two types of data and where more work is needed. We distinguish between five reasons why researchers might want to integrate the two data sources and how this has been helpful: (1) verification, for example, confirming start and end of passively detected trips, (2) contextualization, for example, asking about the purpose of a passively detected trip, (3) quantifying relationships, for example, quantifying the association between self-reported stress and passively measured sleep duration, (4) building composite measures, for example, measuring components of stress that participants are aware of through self-reports and those they are not through passively measured speech attributes, and (5) triggering measurement, for example, asking survey questions contingent on certain passively measured events or participant locations. We discuss challenges of collecting self-reports and passively tracking participants’ behavior with smartphones from the perspective of representation (e.g., who owns a smartphone and who is willing to share their data), measurement (e.g., different levels of temporal granularity in self-reports and passively collected data), and privacy considerations (e.g., the greater intrusiveness of passive measurement than self-reports). While we see real potential in this approach it is not yet clear if its impact will be incremental or will revolutionize the field.
Article
Full-text available
p>This paper examines non-response in a mobile app study designed to collect expenditure data. We invited 2,383 members of the nationally representative Understanding Society Innovation Panel in Great Britain to download an app to record their spending on goods and services: Participants were asked to scan receipts or report spending directly in the app every day for a month. We examine participation at different stages of the process. We further use data from the prior wave of the panel to examine the prevalence of potential barriers to participation, including access, ability and willingness to use different mobile technologies, and biases in the types of people who participate, considering socio-demographic characteristics, financial position and financial behaviours. While the participation rate was low, drop out was also low: Over 80% of participants remained in the study for the full month. The main barriers to participation were access to, and frequency of use of mobile devices, willingness to download an app for a survey, and general cooperativeness with the survey. While there were strong biases in who participated in terms of socio-demographic characteristics (with women, younger, and more educated sample members being more likely to participate), and in terms of financial behaviours (with respondents who already use mobile devices to monitor their finances being more likely to participate), we found no biases in correlates of spending.</p
Article
Full-text available
The rising penetration of smartphones now gives researchers the chance to collect data from smartphone users through passive mobile data collection via apps. Examples of passively collected data include geolocation, physical movements, online behavior and browser history, and app usage. However, to passively collect data from smartphones, participants need to agree to download a research app to their smartphone. This leads to concerns about nonconsent and nonparticipation. In the current study, we assess the circumstances under which smartphone users are willing to participate in passive mobile data collection. We surveyed 1,947 members of a German nonprobability online panel who own a smartphone using vignettes that described hypothetical studies where data are automatically collected by a research app on a participant’s smartphone. The vignettes varied the levels of several dimensions of the hypothetical study, and respondents were asked to rate their willingness to participate in such a study. Willingness to participate in passive mobile data collection is strongly influenced by the incentive promised for study participation but also by other study characteristics (sponsor, duration of data collection period, option to switch off the app) as well as respondent characteristics (privacy and security concerns, smartphone experience).
Article
Full-text available
We asked members of the Understanding Society Innovation Panel about their willingness to participate in various data collection tasks on their mobile devices. We find that stated willingness varies considerably depending on the type of activity involved: respondents are less willing to participate in tasks that involve downloading and installing an app, or where data are collected passively. Stated willingness also varies between smartphones and tablets, and between types of respondents: respondents who report higher concerns about the security of data collected with mobile technologies and those who use their devices less intensively are less willing to participate in mobile data collection tasks.
Article
Full-text available
The increasingly widespread use of mobile phone applications (apps) as research tools and cost-effective means of vast data collection raises new methodological challenges. In recent years, it has become a common practice for scientists to design apps that run only on a single operating system, thereby excluding large numbers of users who use a different operating system. However, empirical evidence investigating any selection biases that might result thereof is scarce. Henceforth, we conducted two studies drawing from a large multi-national (Study 1; N = 1,081) and a German-speaking sample (Study 2; N = 2,438). As such Study 1 compared iOS and Android users across an array of key personality traits (i.e., well-being, self-esteem, willingness to take risks, optimism, pessimism, Dark Triad, and the Big Five). Focusing on Big Five personality traits in a broader scope, in addition to smartphone users, Study 2 also examined users of the main computer operating systems (i.e., Mac OS, Windows). In both studies, very few significant differences were found, all of which were of small or even tiny effect size mostly disappearing after sociodemographics had been controlled for. Taken together, minor differences in personality seem to exist, but they are of small to negligible effect size (ranging from OR = 0.919 to 1.344 (Study 1), np² = .005 to .036 (Study 2), respectively) and may reflect differences in sociodemographic composition, rather than operating system of smartphone users. © 2017 Goetz et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Article
Researchers attempting to survey refugees over time face methodological issues because of the transient nature of the target population. In this article, we examine whether applying smartphone technology could alleviate these issues. We interviewed 529 refugees and afterward invited them to four follow-up mobile web surveys and to install a research app for passive mobile data collection. Our main findings are as follows: First, participation in mobile web surveys declines rapidly and is rather selective with significant coverage and nonresponse biases. Second, we do not find any factor predicting types of smartphone ownership, and only low reading proficiency is significantly correlated with app nonparticipation. However, obtaining sufficiently large samples is challenging—only 5 percent of the eligible refugees installed our app. Third, offering a 30 Euro incentive leads to a statistically insignificant increase in participation in passive mobile data collection.
Article
Although web surveys in which respondents are encouraged to use smartphones have started to emerge, it is still unclear whether they are a promising alternative to traditional web surveys in which most respondents use desktop computers. For sample members to participate in smartphone-based surveys, they need to have access to a smartphone and agree to use it to complete the survey; this raises concerns about coverage and nonresponse, as well as measurement if those who agree to participate have any difficulty using smartphones. In an analysis of data from a smartphone versus desktop (within-subjects) experiment conducted in a probability-based web panel, we compare estimates produced by the smartphone web survey (one condition) and PC web survey (other condition). We estimate mode effects and then examine the extent to which these effects are attributable to coverage, nonresponse, and measurement errors in the smartphone-based survey. While mode effects were generally small, we find that the smartphone web survey produced biased estimates relative to PC web for a subset of survey variables. This was largely due to noncoverage and, to a lesser extent, nonresponse. We find no evidence of measurement effects. Our findings point to the trade-off of the advanced data collection opportunities of smartphones and the potential selection errors that such devices may introduce.
Article
The new European General Data Protection Regulation (GDPR) imposes enhanced requirements on digital data collection. This article reports from a 2018 German nationwide population-based probability app study in which participants were asked through a GDPR compliant consent process to share a series of digital trace data, including geolocation, accelerometer data, phone and text messaging logs, app usage, and access to their address books.With about 4,300 invitees and about 650 participants, we demonstrate (1) people were just as willing to share such extensive digital trace data as they were in studies with far more limited requests; (2) despite being provided more decision-related information, participants hardly differentiated between the different data requests made; and (3) once participants gave consent, they did not tend to revoke it. We also show (4) evidence for a widely-held belief that explanations regarding data collection and data usage are often not read carefully, at least not within the app itself, indicating the need for research and user experience improvement to adequately inform and protect participants. We close with suggestions to the field for creating a seal of approval from professional organizations to help the research community promote the safe use of data.
Article
Challenges to survey data collection have increased the costs of social research via face-to-face surveys so much that it may become extremely difficult for social scientists to continue using these methods. A key drawback to less expensive Internet-based alternatives is the threat of biased results from coverage errors in survey data. The rise of Internet-enabled smartphones presents an opportunity to re-examine the issue of Internet coverage for surveys and its implications for coverage bias. Two questions (on Internet access and smartphone ownership) were added to the National Survey of Family Growth (NSFG), a U.S. national probability survey of women and men age 15–44, using a continuous sample design. We examine 16 quarters (4 years) of data, from September 2012 to August 2016. Overall, we estimate that 82.9% of the target NSFG population has Internet access, and 81.6% has a smartphone. Combined, this means that about 90.7% of U.S. residents age 15–44 have Internet access, via either traditional devices or a smartphone. We find some evidence of compensatory coverage when looking at key race/ethnicity and age subgroups. For instance, while Black teens (15–18) have the lowest estimated rate of Internet access (81.9%) and the lowest rate of smartphone usage (72.6%), an estimated 88.0% of this subgroup has some form of Internet access. We also examine the socio-demographic correlates of Internet and smartphone coverage, separately and combined, as indicators of technology access in this population. In addition, we look at the effect of differential coverage on key estimates produced by the NSFG, related to fertility, family formation, and sexual activity. While this does not address nonresponse or measurement biases that may differ for alternative modes, our paper has implications for possible coverage biases that may arise when switching to a Web-based mode of data collection, either for follow-up surveys or to replace the main face-to-face data collection.