How to Increase Survey Response and Manage the Postal and
Face-to-Face Field Work of Participant Recruitment during the
A Note with Preliminary Evidence from Immigrant-Origin and
Native Voters in a German Metropolis in Spring 2021
Achim Goerres, University of Duisburg-Essen, Corresponding Author, Achim.Goerres@uni-
Jonas Elis, University of Duisburg-Essen
Sabrina J. Mayer, University of Duisburg-Essen and DeZIM Institute Berlin
Dennis C. Spies, Heinrich Heine University Düsseldorf
This Version 1.0, 24 May 2021
This note presents preliminary evidence from a postal and face-to-face recruitment field for a
telephone survey during the Covid-19 pandemic. Drawn from the city register of Duisburg, a
metropolis of 500,000 inhabitants, voters eligible for the September 2021 Bundestag elections
are the target population. They are stratified into four groups by onomastic classification:
native Germans, Germans of Turkish Descent, Germans with a (Post-) Soviet background and
Germans of any other immigrant origin. This note presents ten measures of optimising the
recruitment of individuals from these four groups under the constraints of an all-
encompassing high-incidence pandemic with distance measures and curfews in place, with
closed shops and schools and reduced other care facilities. In addition, the report puts forward
statistical analyses of some measures applied in an experimental design and estimates the
overall and per-person-recruited costs.
Overall, the pandemic forced us to make drastic changes to research design and field
organisation and to invest much more time and money than planned. Especially overburdened
postal deliveries and restrictive policies for public places made the endeavour most difficult.
So far, the ongoing fieldwork has produced a response rate of 8.2 % (only individuals willing
to participate in telephone interview) with 11.0 % as a likely expected outcome. One individual
recruited to take part in the telephone panel survey costs an estimated 53-71 €.
Early-bird incentives produced a positive response (meaning the willingness to partake in the
survey) of 5.5 % compared to 4.0 % among those without early-bird incentives. The most
effective measure to boost overall positive response among initial non-responders was in-
person canvassing with unconditional incentive (21.3 %) instead of postal reminders (9.9 %).
Canvassing a smaller group of people in person proved to be as costly as reminding a larger
group of people in writing: we estimated costs of 60-61 € per additionally recruited survey
participants through either channel.
The project is funded by the German Research Foundation (grant number GO 1833/7-1). We thank Rainer Schnell for advice at
various stages of the project. The Immigrant German Election Study II can be followed online at http://bit.ly/imges2 Please let
us know what you think by sending e-mails to the corresponding author.
Principal Investigators of the grant: AG, SJM, DCS; Research design of the survey: AG, SJM, DCS; Field Management: JE; Design
Contact letters: JE, SJM; Design and Execution of Trainings: AG; Data preparation: JE; Statistical Analysis: AG, JE; Conception
and first draft of the paper: AG; final draft of the paper: AG, JE, SJM, DCS.
This note presents our preliminary findings from collecting survey data from a complex
random sample in a German metropolis in spring 2021 during the pandemic. The objective of
this note is to publish our findings, positive and negative, as quickly as possible in order to help
other survey researchers during the Covid-19 pandemic. Evidence and citations will thus be
minimal. Whereas this first version will include preliminary empirical results: some simple
experiments and some univariate and bivariate observations, later iterations of the note will
include more sophisticated analyses. The survey is still in the field during the time of this note’s
first publication (24 May 2021).
For this study, we have employed the ten measures to recruit participants into our telephone
panel survey between the end of March and the end of May 2021 with the telephone survey
scheduled for June to October 2021. These ten measures do not represent the full area of
decisions to maximise recruitment, but the ones that were most affected by the pandemic.
First, we will present the Immigrant German Election Study II for which the recruitment was
carried out. Then we will discuss each measure in turn, revealing things that have and have not
worked, and estimate, if possible, its causal impact on increasing the willingness to participate
in the survey. Finally, we compare some of the roughly estimated costs.
Overall, the pandemic forced us to make drastic changes to research design and field
organisation and to invest much more time and money than planned. The field work has
produced a response rate of 8.2 % so far. One individual recruited to take part in the telephone
panel survey costs an estimated 53-71 €. Early-bird incentives produced positive response
(meaning the willingness to partake in the survey) of 5.5 % compared to 4.0 % among those
without early-bird incentives. The most effective measure to boost overall positive response
among initial non-responders was in-person canvassing (21.3 %) instead of postal reminders
(9.9 %). Canvassing a smaller group of people in p erson proved to be as expensive as r eminding
a larger group of people in writing: we estimated costs of 60-61 € per additionally recruited
survey participants through either channel.
Description of the Immigrant German Election Study II
The project, in which we have made these experiences, is the Immigrant German Election
Study II (http://bit.ly/imges2 ). It has been financed by the German Research Foundation
since April 2021 and consists of a panel survey of German citizens living in Duisburg, aged 18
and older, at the 2021 Bundestag election that is scheduled for 26 September 2021. The
objective is to follow German voters of immigrant origin and natives through the built-up of
the electoral campaign, the campaign itself and after the election.
Duisburg, a city of 500,000 inhabitants that forms the most Western part of the Ruhr Valley,
an agglomeration of 10 million people in North Rhine Westphalia, Germany’s most populous
state. Duisburg is a very diverse city with 44 % inhabitants who are foreigners or who are first
or second generation immigrant German citizens (Ministerium für Kinder, Familie,
Flüchtlinge und Integration NRW 2020: 20). Duisburg is socially an unequal city with a richer
South bordering with the state capital Düsseldorf and a poorer North. It includes some
suburban areas, especially left of the River Rhine as well as highly densely-populated inner city
For all the time of the project so far, the Covid-19 pandemic was rampant in Duisburg (with an
incidence level above the North Rhine-Westphalian mean) with self-tests but no vaccination
yet available to staff and public life shut down. When we started our fieldwork on 24 March
2021, the 7-days incidence level per 100.000 inhabitants was 157.8. It then steadily rose to a
maximum of 243.6 on 25 April declining afterwards to 66.6 today (24 May) (Corona in Zahlen
2021). Within Duisburg, there were stark differences reflecting the robust correlation between
Covid-19 numbers and social deprivation. For example, in the week 12-18 April the 46
boroughs showed a range of incidence levels between 33.2 (Rahm) and 514.2 (Alt-Hamborn)
with a mean of 186.5 and a standard deviation of 109.2 (Data original from the City of
Duisburg, provided by Infas 360 2021).
Originally, before Covid-19, the research design implied a stratified random sample of
immigrant-origin and native voters from the resident population:
(a) of first and second generation immigrant-origin voters with a Turkish background
(b) first and second generation immigrant origin voters from the Soviet Union or one of its
successor states (Russian Germans) and
(c) voters with no migration background who can be third or higher-order generation
To this end, we requested 70.000 addresses of German citizens born 26 September 2003 or
earlier from the city registration office. By means of onomastic classification, we wanted to
identify individuals as natives, Turkish Germans, Russian Germans or Germans of a different
immigrant background. About 55.000 people were classified as Native Germans, 6.864 as of
Turkish Origin, 1.854 as Russian Germans and 6.848 people as of other immigrant origin.
According to our original plan, we would have drawn sub-samples of three groups, neglecting
the group of Germans of other immigrant backgrounds. We then wanted to interview each
individual four times, a long face-to-face interview after a short info letter announcing that the
interviewers would drop by, in the respondents’ home of about 60 minutes. In that initial
interview, we wanted to collect a lot of time-invariant variables. Finally, we wanted to switch
the survey mode to phone interview and collect time-variant information in three waves in
June – August – October.
Measure 1: Drop face-to-face interviews as the initial survey wave
There was no possibility of doing face-to-face interviews in people’s homes in spring 2021 in
Germany. There were restrictions in place: a limited number of individuals that were allowed
to meet, minimum physical distance of at least 1.5 m, mandatory medical masks in public
transport and supermarkets, closed shops and restaurants except for bare necessities. We thus
had to change our survey design completely and drop the initial face-to-face interview, replace
it with a minimal written survey and move many more questions to the phone interviews.
We thus started with a contact letter inviting people to the survey that included a minimal
initial written survey. This initial survey included questions about where the respondents had
been born in order to assess whether they had been correctly classified. At this point (would
matter less later), it was still possible that a wrongly classified individual could be dropped
from the survey as overcoverage if they were classified to belong to the group of voters of
another immigrant origin
The invitation to be part of the survey could be filled out via paper and pencil and sent back by
pre-paid letter or online by means of a QR code or shortened URL that asked for the entry of a
token. There was a high number of returns by post (~1,700) compared to online (~650) across
Measure 2: Provide a local landline number associated with the university
and contact person for questions
In all letters, we included a Duisburg landline number of the university and university e-mail
address that people could contact if they had any questions. This landline number was called
by approximately 120 respondents in 10 weeks and the call transferred to the mobile of the
field manager Jonas Elis. About 20 of the callers could be converted to take part in the study.
Another 70 callers informed us about a change of residence, death, disability or other kinds of
nonresponse of a contact/target person, the rest were hard rejections. The e-mail address
provided was contacted by 20 individuals in 10 weeks.
Measure 3: Soft-launch of about 12 % of our original sample with
additional test letters and without early-bird incentives
We worked together with the university printing services to create the multiple-page contact
letter including a response envelope. These letters were then sent out to the Duisburg addresses
with the postal provider Postcon. We drew a stratified random sample from our target samples
with about 1,000 persons-registered-at-addresses and send their first-contact letters out early.
Also, we included some additional test letters to work colleagues not in our sample. By this, we
wanted to test the various complex organisational processes, printing and posting, and see
when the letters would arrive and in what state they would be in.
This soft launch revealed a major problem with the delivery. Recall that the letters were sent
out from a Duisburg address only to Duisburg addresses. This would normally require one
working day maximum. However, it took Postcon more than one week to deliver the first 1,000
letters (and we do not have 100 % proof that they sent out all). Due to the security measure of
the additional test letters, we quickly learnt about the problem and could escalate the problem
with Postcon. Their internal investigation found that their delivery infrastructure for some
parts of Duisburg had failed due to illnesses (possibly related to Covid-19). Recall that this was
a period of increased postal circulation because people ordered more online, and small-scale
deliveries would go through the mail system. Throughout the postal process, we subsequently
had a level of uncertainty about deliveries and their dates.
We then sent out the full launch, totalling 3,000 individuals classified as natives, 3,000
individuals classified as Turkish Germans and all Russian Germans from our sample (about
1,700 in total), including an early-bird incentive.
Measure 4: Early-bird incentive
Of those treated with an early-bird incentive (a 5 € gift card from Amazon) for getting back to
us by a set date about two weeks after the posting of the letters, 550 out of 7700 (7.1 %)
individuals responded, among which 420 with the willingness to participate (5.5 %). This is
significantly higher than the response rate of 4.0 % among those who had not received the
promise of an early bird-incentive (p=.023 of a one-sided test). This is not much of a difference
and its small size could have been due to problems with postal delivery. Apparently, some
people received their letter one day before the deadline of the early-bird incentive. Postcon,
normally a reliable provider for commercial services, had been severely hampered by Covid-19
related consequences on its staff. The early-bird incentive, normally an efficient way of
increasing response, had not fully delivered. However, it remains a relatively cheap measure,
even when hampered by unreliable postal delivery, as the cost estimates further down will
Measure 5: Reminding by letter and by postcards
As is common practice, we reminded all target individuals in writing except for a sub-sample
of about 2,000 individuals who were selected for in-person canvassing. The first reminder was
a similar letter with response envelope to the initial letter. The second reminder was simply a
postcard. Reminders were sent out if we had not heard anything from the target person.
Table 1: Returns for postal reminders
Group Number of target
had not answered
by the time of the
% of targeted who
All individuals 6,369 700 11.0
Natives 3,157 438 13.9
Turkish Germans 2,406 184 8.0
Russian Germans 806 78 9.7
Not available yet
The first reminder proved to be successful and brought additionally 700 responses. This
represents 11.0 % of all respondents who received a postal reminder. Almost all responses (9.9
%) were positive. i.e. the individuals agreed to participate in the following surveys. As expected,
the response rates were significantly lower among immigrant-origin individuals with the
difference between the two groups too small, too be statistically significant (p=0.132 of a two-
Individuals who received postal reminders were all individuals classified as having no
migration background. For the two groups of immigrant-origin voters, we drew a random sub-
sample from all those individuals who had not answered yet. The other part of this group
received in-person canvassing. We will present our analysis of this difference in modes in the
Measure 6: In-person canvassing with unconditional incentive among an
We were expecting a low response rate because of the pandemic. People spent more time at
home, but they were more stressed, some had less time because of massively increased care
responsibilities with children being home-schooled in full or partially during that period. We
saw a particularly low response rate among the Turkish Germans and Russian Germans after
the first weeks that we had also expected. We thus decided to design a strategy to safely canvass
respondents whom we had not heard from at their doorstep, train a group of interviewers and
equip them in time with the necessary material.
We decided to visit each person only once (because these visits were so expensive), and to leave
a personalised note from the interviewer (“Achim Goerres visited you on 12 April at 14:30 and
could not talk to anyone from your household”) whenever the interviewer met no one in the
household. Also, we left the same reminding material that was sent out to all other people
either directly with the target person or the contact person and an unconditional incentive of
5 € in the form of a bill. We focused our endeavours with a few exceptions on Turkish Germans
and Russian Germans. Interviewers filled out a contact protocol online through their
smartphone after each visit using tokens for the target person.
In total, 17 interviewers were trained to canvass during this hot phase of the pandemic. The
interviewers consisted of three researchers and 14 student assistants. The student assistants
were either employed by the universities on hourly contracts or received a project-based one-
Spies, Elis and Goerres, who were based locally, went first into the field to canvass for 1 full day
to 2.5 full days each. We did this in order to optimise some of the routines, such as improving
the online protocol, and in order to lead by example. We wanted to show those working for us
that we were exposing us to the same risks and risk-mitigation protocols while canvassing as
The training consisted of four hours of online training in the evening and four hours of face-
to-face training on campus. The face-to-face training was scheduled for the end of March 2021.
At that time, all social meetings were prohibited. We could legally meet only outside, on
university premises for a work-related meeting that could not be replaced by a virtual meeting.
We designed a set of 26 vignettes with all potential constellations of meeting and not meeting
target persons in the field. Canvassers practiced in groups of three with special attention to
dialogues through the intercom without face-to-face interaction and unusual situation arising
from the pandemic.
We had no blueprint as to managing fieldwork during a pandemic. Most importantly, we were
uncertain about the risks of talking to people face-to-face in doorways and staircases. On 18
May 2021, after our canvassing operation had been finished, there was a pandemic break-out
of 26 infected people due to sharing the elevator in sequence in a high-riser in Velbert near-by
(WDR 2021). This event demonstrates how real the danger of infection in halls and doorways
We were also uncertain about the development of the pandemic and the variation of that
danger across Duisburg. We thus decided for the following set of security measures against the
pandemic itself and its psychological consequences:
canvassers had to wear masks on public transport and whenever indoors in doorways or
canvassers were not allowed to enter into individual apartments
canvassers had to self-test before starting a canvassing stint
canvassers had to disinfect their hands in regular intervals
canvassers were trained about how to deal with aggressive behaviour
canvassers were asked to contact the field manager in case of danger or any questions
canvassers and the project team shared an instant messaging group to share experiences
Canvassers received batches of 20 persons-registered-at-addresses that were in the same
postcode area. They received no digital transmission of personal data and had to pick up these
batches from university premises. They were also provided with the 20 addresses, without
names, digitally, so that they could plan their trips in an efficient way. They had been shown
how to use a freely accessibly online platform that allowed addresses to be imported from
digital lists and flagged on digital maps. They were required to delete all these addresses after
finishing their tours.1
The composition of batches per interviewer followed the logic of equal risk-sharing. We knew
that the Covid-19 incidence rates in Duisburg varied massively by socio-economic composition
of the boroughs (unemployment rate from 2011 census and Covid-19 incidence rates from 10
March 2021). We thus tried to equip each interview with a mix of batches from low- and high-
SES areas. To that end, we compiled an index per postcode of social deprivation.
In-person canvassing proved to be highly effective (see Table 2).
1 Canvassers were also provided with carefully researched publicly accessible bathrooms since all bars,
shops, restaurants and libraries were closed. Most of these toilets proved to be closed, too. While in the
field, accessible bathrooms at gas stations and cemeteries were identified.
Table 2: Returns for in-person canvassing
Group Number of target
had not answered
% of targeted who
All individuals 2,010 610 30.3
Natives 70 36 51.0
Turkish Germans 970 276 28.0
Russian Germans 970 298 30.7
A response means the target person has filled out the contact questionnaire online or via
returning post. It does not mean the canvasser spoke to the target person.
We concentrated our canvassing efforts on the two main groups of immigrant-origin voters.
Only 70 natives were canvassed, but producing a response of 51.0 %. Among 970 Turkish
Germans, we reached a response rate of 28.0 %, and from 970 Russian Germans a response
rate of 30.7 %. The general response rate was 30.3 %, only counting positive responses,
willingness to take part, lowers the rate to 21.3 %.
Since the canvassed group was randomly assigned, we can compare it with the other group that
received only written reminders. We can estimate the difference in effectiveness and compare
the 11.0 % response from postal reminders with 30.3 % response of in-person-canvassing.
Obviously, this difference of 19.3 percentage points is significant (two-sample t-test, p<0.001)
with the following 95%-confidence interval of difference: [17.6; 21.1].
For the positive responses the difference between canvassing (21.3 %) and postal reminders
only (9.9 %) is also highly significant and tips in favour of canvassing. The 95%-confidence
interval of the difference in positive response is [9.8 %; 13.1 %].
Given that in-person canvassing was focused on the harder-to-reach populations, the estimates
are underrating the size of the added value of canvassing.
Future iteration of the paper will provide estimates about the effectiveness of canvassing as a
function of interviewer qualities, Covid-19 numbers and socio-economic indicators at borough-
Measure 7: Registration of a time-stamped set of decision rules to decide
by in case of sudden changes during the field period
In order to be able to change the field management guidelines in case of an explosion of Covid-
19 case numbers or any other unforeseen changes, the three principal investigators Goerres,
Mayer, and Spies agreed on a set of deciding guidelines that would allow for quick decisions
taken by the collective. This set of principles was uploaded and time-stamped on the Open
Science Framework https://osf.io/t3mp6 . This precaution was taken as our decision might
theoretically had meant endangering human beings and we thus wanted to enable our future
selves under pressure and in unknown territory to decide under a fixed set of decision rules. In
addition, the study and all measures were talked through with the safety specialists at the
University of Duisburg-Essen. Fortunately, these decision rules never had to be invoked. All
canvassers finished their patches unharmed and with no infection.
Measure 8: Topping up stratified samples
After 4 weeks in the field, we topped up samples of Turkish Germans by 3,364 (the rest of all
those classified as of Turkish origin) and Native Germans by 1,000 people (randomly drawn)
in order to increase the absolute number of respondents from these groups. For the Russian
Germans, recall that we had already employed the full sample from onomastically classified
Measure 9: Collaboration with the local media
After three days in the field, we initiated contact with the local newspaper with the highest
popularity and asked them whether they wanted to cover our canvassing efforts, which they
did on 19 April (Ahlers 2021). The coverage was intended to increase the social benefit of
participation in our survey. We hypothesised that target individuals were more likely to
participate in a study that they or those close to them had read or heard about.
It is impossible to judge the impact of that publication given that it happened simultaneously
with the beginning of the roll-out of the canvassing operation. The newspaper report was read
~460 times2, but the print circulation of the WAZ was about 32,000 in Duisburg. It was
subsequently noticed by the local public TV channel Westdeutscher Rundfunk who created
their own 5-minute story for the local news broadcast that aired on 10 May. We only have
anecdotal evidence in the form of single respondents mentioning the TV report.
Measure 10: Add another stratum of sample
As the response rates for Turkish Germans and Russian Germans were still below 10 % after
four weeks in the field, we mobilised another sample from our onomastically classified group
of individuals, namely immigrant-origin Germans who are neither Turkish Germans nor
Russian Germans. This was intended to increase the absolute size of our panel study. Given
the expected small change effect sizes across time and the unknown levels of panel mortality
during a pandemic, we wanted to make sure that we would have enough statistical power to
estimate overtime changes with sufficient precision. The recruitment data for that group are
not available yet.
This strategic decision also had the advantage that we no longer had overcoverage individuals.
All German citizens of the right age were now part of the target population. A target person
wrongly classified as a Russian German who turned out to be a native could be added to the
other stratum. Post-hoc weighing now allows us to correct for differences in sampling
Cost effectiveness of the measures
For managing the field, we employed resources in terms of time and money. In comparison,
we can now look at the cost efficiency of the various measures. This is not a straightforward
exercise because some of the measures cross-cut and because we have not fully finished
collecting all data.
In total, the collection of respondents willing to participate in the survey was about 116,836 €.
So far, this generated 1,650 individuals with full details and willingness to be interviewed in
the telephone panel survey. This sums up to 70.81 € per person, including the labour costs of
the field manager and student assistants and excluding the labour costs of the principal
investigators. This number is likely to decline as we are still expecting further positive
2 Personal communication with the WAZ team, 20 May 2021.
responses. We believe that 2,200 willing participants is a likely maximum, meaning that 53.11
€ is an optimistic lower boundary of that corridor.
We will thus focus on early-bird incentives, postal reminders and in-person canvassing. All of
these measures were applied to random sub-samples, so that causal effects can be estimated.
Table 3 reveals the input in terms of field manager and student assistant time (neglecting PI
time) and monetary costs. This is all on top of the costs of the “normal” first postal contact that
we neglect here as it is the same for everyone. As a reading example: 7,700 people received a
first-contact letter with the promise of an early-bird incentive. Out of that group 550 people
answered to the early-bird incentive, 420 of which were positive, meaning they were willing to
participate in the survey.
We can thus calculate the additional costs of one additional positive response. Giving no early-
bird incentive only produces additional 2.7 € “handling” costs per positive response. Early-bird
incentives increase the costs to 4.79 €. Sending out postal reminders to a lot of people costs
almost exactly the same per additionally recruited individual (60.36 €) as canvassing a smaller
groups with higher return (60.91 €).
This short note presented preliminary evidence from our field operations of recruiting voters
in a German metropolis for a telephone panel survey during the Covid-19 pandemic in March-
May 2021. The pandemic severely limited us in our possibilities due to the harsh constraints
on movement, distancing measures and overburdened postal services.
So far, the ongoing field work has produced a response rate of 8.2 % (only those willing to
participate) with 11.0 % a likely outcome considering that we still are in the field. One
individual recruited to take part in the telephone panel survey costs an estimated 53-71 € with
especially the postal costs and canvassing being expensive.
Early-bird incentives produced positive response of 5.5 % compared to 4.0 % among those
without early-bird incentive. The most effective measure to boost overall positive response
among initial non-responders was in-person canvassing (21.3 %) instead of postal reminders
(9.9 %). Canvassing a smaller group of people proved to be as expensive as reminding a larger
group of people with an estimated 60-61 € costs per additionally recruited positive response
through either channel.
All evidence is preliminary and subject to change. But we see already how seemingly expensive
canvassing measures to reach hard-to-reach respondents creates high returns. These returns
of canvassing a few are, as to costs, not different from the less efficient postal reminders that
are sent out to many people.
Table 3: Three measures in comparison as to their costs
costs in €
costs in €
Early-bird incentives in full launch 4 0 1,905 2,013 7,700 550 420 4.79
No early-bird incentive in soft launch 4 0 0 108 1,000 50 40 2.70
Reminders by letters and postcards 10 0 45,000 45,270 17,200 1050 750 60.36
80 850 12,500 27,410 2,010 636 450 60.91
Assumed gross hourly costs student assistants 15 €, field manager 27 €. All costs are estimated. Basic costs of sending out one letter to each individuals
are not listed as this was incurred for every individual.
Ahlers, Martin (2021): Wahlstudie: Duisburger Politik-Professor auf Klapprad-Runde, 19 April
2021, available at https://www.waz.de/staedte/duisburg/wahlstudie-duisburger-politik-
Corona in Zahlen (2021): Kresifreie Stadt Duisburg, 24 May 2021, https://www.corona-in-
Infas 360 (2021): Corona-Datenplattform, available at https://www.corona-
datenplattform.de/, data provided on 5 May 2021.
Ministerium für Kinder, Familie, Flüchtlinge und Integration NRW (2020): Integrationsprofil
Duisburg. Daten zu Zuwanderung und Integration, available at
WDR (2021): Ticker vom Samstag (22.05.2021) zum Nachlesen, available at