ArticlePDF Available

The Language Experience and Proficiency Questionnaire (LEAP-Q): Assessing Language Profiles in Bilinguals and Multilinguals

Authors:

Abstract and Figures

To develop a reliable and valid questionnaire of bilingual language status with predictable relationships between self-reported and behavioral measures. In Study 1, the internal validity of the Language Experience and Proficiency Questionnaire (LEAP-Q) was established on the basis of self-reported data from 52 multilingual adult participants. In Study 2, criterion-based validity was established on the basis of standardized language tests and self-reported measures from 50 adult Spanish-English bilinguals. Reliability and validity of the questionnaire were established on healthy adults whose literacy levels were equivalent to that of someone with a high school education or higher. Factor analyses revealed consistent factors across both studies and suggested that the LEAP-Q was internally valid. Multiple regression and correlation analyses established criterion-based validity and suggested that self-reports were reliable indicators of language performance. Self-reported reading proficiency was a more accurate predictor of first-language performance, and self-reported speaking proficiency was a more accurate predictor of second-language performance. Although global measures of self-reported proficiency were generally predictive of language ability, deriving a precise estimate of performance on a particular task required that specific aspects of language history be taken into account. The LEAP-Q is a valid, reliable, and efficient tool for assessing the language profiles of multilingual, neurologically intact adult populations in research settings.
Content may be subject to copyright.
The Language Experience and
Proficiency Questionnaire (LEAP-Q):
Assessing Language Profiles in
Bilinguals and Multilinguals
Purpose: To develop a reliable and valid questionnaire of bilingual language status
with predictable relationships between self-reported and behavioral measures.
Method: In Study 1, the internal validity of the Language Experience and Proficiency
Questionnaire (LEAP-Q) was established on the basis of self-reported data from
52 multilingual adult participants. In Study 2, criterion-based validity was established
on the basis of standardized language tests and self-reported measures from 50 adult
SpanishEnglish bilinguals. Reliability and validity of the questionnaire were
established on healthy adults whose literacy levels were equivalent to that of someone
with a high school education or higher.
Results: Factor analyses revealed consistent factors across both studies and suggested
that the LEAP-Q was internally valid. Multiple regression and correlation analyses
established criterion-based validity and suggested that self-reports were reliable
indicators of language performance. Self-reported reading proficiency was a more
accurate predictor of first-language performance, and self-reported speaking proficiency
was a more accurate predictor of second-language performance. Although global
measures of self-reported proficiency were generally predictive of language ability,
deriving a precise estimate of performance on a particular task required that specific
aspects of language history be taken into account.
Conclusion: The LEAP-Q is a valid, reliable, and efficient tool for assessing the
language profiles of multilingual, neurologically intact adult populations in research
settings.
KEY WORDS: bilingualism, self-assessment, second language,
language proficiency, questionnaire development
B
ilingualism and multilingualism are the norm rather than the excep-
tion in todays world (Harris & McGhee-Nelson, 1992), and the pro-
portion of linguistically diverse populations is increasing in the United
States ( U.S. Bureau of the Census, 2003). These demographic changes are
reflected in the growing representation of multilingual and multicultural
populations in research and applied settings. However, research with bilin-
guals often yields inconsistent findings (e.g., Grosjean, 2004; Marian, in
press; Romaine, 1995). For example, bilingual cortical organization (e.g.,
Kim, Relkin, Lee, & Hirsch, 1997; Marian, Spivey, & Hirsch, 2003; Perani
et al., 1998; Vaid & Hull, 2002), lexical processing (e.g., Chapnik-Smith,
1997; Chen, 1992; Kroll & de Groot, 1997), and phonological and ortho-
graphic processing (e.g., Doctor & Klein, 1992; Grainger, 1993; Macnamara
& Kushnir, 1971; Marian & Spivey, 2003) have all been found to differ
Viorica Marian
Henrike K. Blumenfeld
Margarita Kaushanskaya
Northwestern University, Evanston, IL
Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007 D American Speech-Language-Hearing Association
1092-4388/07/5004-0940
940
depending on bilinguals ages of language acquisition,
mode(s) of acquisition, history of use, and degree of pro-
ficiency and dominance. These inconsistencies are further
exacerbated by the absence of uniform assessment instru-
ments in bilingualism research. Those who work with bi-
linguals and multilinguals often fac e th e challenge of
testing individuals whose language they do not speak
(Roseberry-McKibbin, Brice, & O Hanlon, 2005) and thus
have to rely exclusively on self-assessed information, usu-
ally collected with improvised questionnaires. The need
for a language self-assessment tool that is comprehen-
sive, valid, and reliable across bilingual populations and
settings prompted a systematic approach to developing
the present Language Experience and Proficiency Ques-
tionnaire (LEAP-Q; see Appendix).
Previous Self-Assessment Studies
In general, previous research suggests that self-
reported language measures are indicative of linguistic
ability (e.g., Bachman & Palmar, 1985; MacIntyre, Noels,
& Clement, 1997; Ross, 1998; Shameem, 1998; Stefani,
1994). Existing se lf-assessment tools for studying bilin-
guals span both domain-general ( e.g., B ahrick, Hall,
Goggin,Bahrick,&Berger,1994; Delgado, Guerrero,
Goggin, & Ellis, 1999) and domain-specific proficiency
(e.g., Flege, Yeni-Komishian, & Liu, 1999; Jia, Aaronson,
& Wu, 2002; Vaid & Menon, 2000). For instance, in a
study of the relation ship betwee n self-reported profi-
ciency and language performance, Delgado et al. (1999)
tested SpanishEnglish bilinguals and correlated self-
assessed proficiency i n English and Spanish with per-
formance on the WoodcockMuñoz Language Survey
(Woodcock & Muñoz-Sandoval, 1993). Delgad o et al.
found that participants assessed first-language (L1)
skills more accurately than they did second-langua ge (L2)
skills. WoodcockMuñoz scores correlated with all self-
reported measures of L1 proficiency but with only s elf-
reported measures of L2 reading and writing (and not
with L2 speaking and understanding). Similarly, Bahrick
et al. (1994) found that language dominance ratings corre-
lated highly with performance on some tasks (e.g., category
generation and vocabulary recognition) but correlated less
with performance on other tasks (e.g., oral comprehension).
Together, studies of domain-general self-assesement in
bilinguals suggest that the relationship between self-
reported and behavioral measures of language perfor-
mance varies across languages and tasks (e.g., Bahrick
et al., 1994; Delgado et al., 1999).
Studies of domain-specific proficiency assessment in
bilinguals have focused on grammatical ability (e.g., Jia
et al., 2002), the degree of foreign accent (e.g., Flege et al.,
1999), and computational language use (e.g., Vaid
& Menon, 2000). For instance, Jia et al. us ed a 32-item
questionnaire that assessed the following four areas:
(a) age/time variables associated with L2 acquisition;
(b) environmental variables (e.g., number of L2 speakers
at home; frequency with which L2 is spoken at home and
in the workplace; and fathers, mother s, and siblings pro-
ficiency in speaking, reading, and writing L2); (c) affective
variables (e.g., self-consiousness, cultural preference and
identity, and motivation); and (d) self-evaluated L1 and L2
proficiency in speaking, reading, and writing. Jia at al.
found that self-reported ratings of language proficiency
were positively correlated with behavioral performance.
Similarly, in a series of studies on how language domi-
nanceaffectsthedegreeofforeign accent and grammatical
ability, Flege, MacKay, and Piske (2002) and Flege
et al. (1999) used a language history questionnaire that
targeted participants self-reported age of arrival in the
L2-speaking country/ initial L2 learning; age of attained
L2 proficiency; duration of L2 immersion; nu mber of years
of L2 schooling; percentage use of L1/L2; frequency of ex-
posure to L2 TV, movies/videos, and radio; frequency of
use of L1/L2 in a working environment; and ability to
imitate foreign accents. Flege et al. (1999) found signif-
icant correlations between language history and degree of
foreign accent in L2 and between language history and
performance on a grammaticality judgment task (Flege
et al., 2002). In a study of compuational language use,
Vaid and Menon (2000) used a questionnaire that focused
on language preference for mental arithmetic (e.g., count-
ing, noting the time, remembering a telephone number).
The questionnaire also yielded self-reported ratings of
language proficiency speaking, comprehending, reading,
and writing, as well as participants age of arrival in the
L2-speaking country/initial L2 learning; the setting of
language acquisition (home, school, other); duration of L2
immersion; the language of instruction in elementary and
secondary school; and frequency of use of L1/L2 at work,
with parents, and with siblings, while thinking to self,
and while dreaming. Lan guage preference f or mental
arithmetic was found to correlate with variables in the
bilinguals language history, with the strongest predictor
being the language of early formal instruction followed by
length of residence in the L2 country, onset of bilingual-
ism, and relative language dominance.
A consistent aspect of these studies was their focus
on proficiency and history-related variables. However,
the studies diverged in three notable ways: (a) The dis-
tinction among language proficiency, dominance, and pre-
ference remained largely unexplored, (b) behavioral tasks
used to validate the questionnaires were limited, and
(c) questions and scales were not consistent across stud-
ies. There is currently no uniform procedure for deter-
mining bilingual language dominance and proficiency.
Researchers frequently use distinct aspects of language
status and performance to delineate the two, or they use
the same measure (e.g., L1:L2 proficiency ratio) to define
Marian et al.: Bilingual LEAP Questionnaire 941
both dominance (e.g., Flege et al., 2002) and proficiency
(e.g., Vaid & Menon, 2000). For instance, whereas some
studies have used self-ratings of ability to speak, under-
stand, read, and write L1 and L2 (e.g., Goggin, Estrada,
& Villarreal, 1994; Lemmon & Goggin, 1989; Magiste,
1979), others have relied on the experimenterssubjec-
tive judgment to determine dominance in a language
(e.g., Talamas, Krol l, & Dufour, 1999). Factors related
to bilinguals language history, such as language of ex-
posure in early years (e.g., Chincotta & Underwood, 1998;
Hazan & Boulakia , 1993), current language use (e.g.,
Grosjean, 1982), bilinguals speed of executing instructions
(Lambert, 1955), and speed of naming pictures (Magiste,
1992), have all been u sed to operationalize language
dominance. Whereas some researchers have assessed self-
reported proficiency in comprehending, speaking, read-
ing, and writing (e.g., Vaid & Menon, 2000), others have
assessed only some, but not other, proficiency domains
(e.g., Jia et al., 2002, who did not assess proficiency com-
prehending). Similarly, whereas some researchers have
assigned bilinguals into dominance groups (L1 vs. L2)
on t he basis of vocabulary test scores in L1 an d L2
(e.g., Cromdal, 1999), others have done so on the basis
of speed of reading in the two languages (e.g., Favreau
& Segalowitz, 1982; Macnamara, 1969). Adding to the
confusion is that researchers have also, at times, used
language preference, instead of language proficiency or
dominance, as the domain of interest (e.g., Marian &
Neisser, 2000; Ortiz & Garcia, 1990). Language prefer-
ence is typically used to index participants subjective
feelings toward a language. In sum, some studies have
focusedondominanceratingsasindicatorsof truelinguis-
tic performance, but others have focused on proficiency
or on preference ratings.
The second point of divergence among previous stud-
ies stems from ratings of language proficiency that were
often compared to bilinguals performance on only one
or two behavioral tasks, such as degree of foreign accent
(e.g., Flege et al., 1999) or grammaticality judgment (e.g.,
Jia et al., 2002), rather than on a range of behavioral
tasks. There is considerable evidence that various lan-
guage history variables apply differently across perfor-
mance domains: For example, age of acquisition applies
more to phonology but less to morphosyntax (Snow &
Hoefnagel-Höhle, 1979). Therefore, an accurate picture of
the relationship between bilinguals self-reports and per-
formance can be gained only from a comprehensive assess-
ment of behavioral language performance.
Third, although some variables were assessed uni-
formly across studies (e.g., age of L2 learning), others
were research specific (e.g., number of L1 and L2 speak-
ers at home). Researchers target self-reported informa-
tion that is relevant to the experimental manipulation
at handfor example, the ability to imitate a foreign ac-
cent is relevant to studying accents but is less relevant
to studying c omputational language use. The different
experimental goals often result in distinct questionnaires,
making cross-experimental comparisons difficult. Finally ,
the scales used across studies often differ as well. For in-
stance, whereas Jia et al. (2002) used a 4- or 5-point rat-
ing scale for all of their questions, Vaid and Menon (2000)
elicited self-reported ratings on a 7-point scale.
Previous studies suggest that bilinguals language
profiles are best captured by assessing language experi-
ence and proficiency across multiple linguistic domains.
It appears that bilinguals are able to assess their language
proficiency and report their language history in a way that
is consistent with behavioral performance (e.g., Chincotta
& Underwood, 1998; Flege et al., 1999, 2002; Jia et al.,
2002). However, the absence of a valid and uniformly used
assessment measure makes it difficult to interpret exist-
ing findings and to make generalizations across studies
and populations. In this article, we introduce a self-
assessment tool that combines relevant proficiency and
experience variables into a single instrument. We exam-
ine the relative value of each history variable for indexing
bilinguals actual linguistic performance across a range
of standardized behavioral measures.
The Language Experience and
Proficiency Questionnaire
The goal of this project was to develop a reliable and
valid questionnaire for efficient assessment of bilinguals
linguistic profiles. Research indicates that ratings of pro-
ficiency alone are not sufficient to determine bilingual
language status and that bilinguals language learning
and language use experiences play a significant role in
shaping their linguistic competence (e.g., Grosjean, 2004;
Hyltenstam & Abrahamsson, 2003). Therefore, the Lan-
guage Experience and Proficiency Questionnaire (LEAP-Q)
was constructed within the context of bilingualism theo-
ries that view L2 acquisition as an interplay between pro-
ficiency and experience variables (e.g., Hyltenstam &
Abrahamsson, 2003). For exampl e, Flege, Frieda, a nd
Nozawa (1997) and Flege, Frieda, Walley, & Randazza
(1998) demonstrated that bilinguals dominance ratings
were less predictive of L2 performance than experience-
related variables and that although both language dom-
inance and language history accounted for considerable
variance in behavioral performance, language history
variableswerebetterpredictorsofL2performance
than dominance ratings (e.g., the frequency with which
bilinguals spoke L1 influenced their performance in pro-
cessing L2 phonology). It has been proposed that L2 ex-
perience variables become more important in shaping
proficiency with increased L2 acquisition age ( Hyltenstam
& Abrahamsson, 2003; see also de Houwer, 1995) and
that L2 acquisition is a result of cognitive, social, and
942 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
environmental factors (e.g., Bialystok & Hakut a, 1994,
1999; Snow , 1983; Snow & Hoefnagel-Höhle, 1978). There-
fore, in this study, both language proficiency and language
history variables were considered necessary for specifying
bilingual language status.
Given a theoretical framework that incorporates both
language proficiency and language history, the LEAP-Q
aims to capture factors that previously have been iden-
tified as important contributors to bilingual status: lan-
guage competence (including proficiency, dominance, and
preference ratings); age of language acquisition; modes
of language acquisition; prior language exposure; and cur-
rent language use. The LEA P-Q is based on question types
previously used in questionnaires assessing bilinguals
(e.g., Flege et al., 1999, 2002; Jia et al., 2002; Marian &
Spivey, 2003; Vaid & Menon, 2000).
Language competence. Traditionally, the self-
assessment literature has used three distinct measures
to index bilingual language competence: (a) language pro-
ficiency, (b) language dominance, and (c) language pref-
erence. Because conflating the three measures can render
interpretation of results difficult, each of them was
probed separately in the LEAP-Q. Previous studies have
construed profic iency as an index of general abilities
across language proc essing domains ( e.g., Bachman &
Palmar, 1985; Stefani, 1994), including literacy-oriented
proficiency, grammatical proficiency, vocabulary knowl-
edge, and discourse abilities (Bachman, 1990; Harley,
Cummins, Swain, & Allen, 1990). Consistent with other
studies of bilingual self-assessment (e.g., Bahrick et al.,
1994; Flege et al., 1999, 2002; Grosjean, 2004; Jia et al.,
2002; Vaid & Menon, 2000), the LEAP-Q elicited profi-
ciency ratings in speaking, listening, reading, and writing.
However , instead of collapsing proficiency ratings along
the different performance domains into a cumulative
score (e.g. , Flege et al., 2002), profic iency ratings ob-
tained in this study were analyzed separately and were
expected to yield different predictive information for dif-
ferent linguistic skills. For language dominance, partici-
pants in this study indicated dominance order for each of
the languages spoken. The debate around the utility of a
single global measure, such as language dominance (e.g.,
Oyama, 1978; Spolsky, Sigurd, Sako, W alker, & Arterburn,
1968), versus multiple task-specific measures, such as
linguistic proficiency across domains (e.g., Bahrick et al.,
1994; Fishman & Cooper, 1969), prompted the inclusion of
both global (dominance) and specific (proficiency) mea-
sures of language competence, making it possible for the
LEAP-Q to examine their effectiveness in indexing actual
linguistic skills. Finally, LEAP-Q questions targeting pref-
erence were posed in specific terms (e.g., preference regard -
ing reading a text available in all languages) rather than as
general questions about t he over all p referred language
to maximize reliability and interpretation. Although a
comparison of the three measures (proficiency , dominance,
and preference) was not within the s cope of this study , their
availability in the LEAP-Q enables questionnaire users
to weigh each of these measures against their variable of
interest.
Language acquisition. Age of acquisition has been
shown to be tightly connected to language learning, to
influence bilinguals ratings of language dominance, and
to predict their performance o n behavioral t asks (e.g.,
Hyltenstam & Abrahamsson, 2003; Johnson & Newport,
1989). For example, Flege et al. (2002) found that age
of acquisition influenced bilinguals dominance classifi-
cation and correlated with bilinguals sentence duration
ratios in both languages. Consistent with studies demon-
strating maturation effects in L2 acquisition, the LEAP-Q
elicited four age-of-acquisition measures for each language
spoken: (a) age of initial language learning, (b) age of at-
tained fluency, (c) age of initial reading (i.e., age at which
participants started to read in each language), and (d) age
of attained reading fluency.
Moreover, the environment in which a language is
learned also influences proficiency attainment. For in-
stance, Flege et al. (1999) found that the number of years
of education received in an L2 country, years of residence
in an L2 country, average self-estimated use of L1 and
L2,andchronologicalageallinfluenced age-of-acquisition
effects on bilingual language dominance. The importance
of environmental and contextual variables in language
acquisition was demonstrated by Carroll (1967), who found
a significant relationship between language performance
and the extent to which the target language was used in
the home. Therefore, the LEAP-Q elicits descriptions of
acquisition modes in terms of the learning environments
and in terms of the extent to which these learning envi-
ronments contributed to language acquisition.
Prior and current language exposure. The degree of
prior exposure to a language has been shown to influence
research findings (e.g., Birdsong, 2005; Genesee, 1985;
Kohnert, Bates, & Hernandez, 1999; MacKay & Flege,
2004; McDonald, 2000; Weber-Fox & Neville, 1999). For
example, Flege et al. (1999) found that length of residence
in the United States influenced bilinguals sentence-level
performance, with various language abilities differen -
tially susceptible to language exposure (e.g., groups with
more U.S. education had significantly higher rule-based
morphosyntax scores than groups with less U.S. educa-
tion but showed no differences in foreign accent ratings
or in lexically based morphosyntax scores). Given the evi-
dence that prior language exposure influences bilingual
performance, the LEAP-Q assessed exposure to a language
in four different environments: (a) in a country , (b) at school,
(c) at work, and (d) at home.
In addition to prior exposure, ongoing language use
can influence research findings. For example, Jia at al.
(2002) found that mothers L2 proficiency and frequency of
speaking L2 at home were predictive of bilingual childrens
Marian et al.: Bilingual LEAP Questionnaire 943
behavioral performance. Similarly, bilinguals who used L2
more often than L1 had better pronunciation and higher
morphonosyntactic performance in L2 than bilinguals who
used L1 more often than L2 ( Flege et al., 2002). Therefore,
the LEAP-Q elicited information regarding bilinguals cur-
rent exposure to their languages across settings, including
interaction with family and friends and exposure during
reading, watching TV, and listening to the radio, as well as
exposure through self-instruction and language tapes.
LEAP-Q Aims
The target population for the LEAP-Q consists of
adult and adolescent bilinguals and multilinguals with
a variety of language experiences and proficiency levels.
It includes simultaneous bilinguals (who learned their
L1 and L2 early on and in parallel), late bilinguals (who
learned their L2 later in life), balanced bilinguals (who
are equally proficient across their languages), and unbal-
anced bilinguals (who are more proficient in one language
than the other). In its current form, the LEAP-Q has been
validated for use with individuals who have attained lit-
eracy skills equivalent to a high school education level or
higher in at least one of their languages. Although pre-
vious studies suggest that similar questions about pr ofi-
ciency and language history can be successfully used to
capture language profiles in bilingual children by means
of parent reports (Chincotta & Underwood, 1998; Flege
et al., 2002; Vaid & Menon, 2000), the questionnaire has
not yet been validated for use with children or with clin-
ical populations. Finally, unlike language testing instru-
ments in which the primary objective is to place students
in English-as-a-second language and foreign language
programs (e.g., those offered by the Center for Advanced
Research on Language Acquisition and the American
Council on the Teaching of Foreign Languages, as well as
the University of Ontarios French Immersion Program
Assessment Tool), the target settings f or the LEAP-Q
are primarily research oriented (i.e., for assessment of re-
search participants). Although novice language-learners
may complete the questionnaire, the current study vali-
dated only self-reports of bilinguals with proficiency lev-
els sufficient to complete standardized assessment tools
(i.e., for self-reports to be meaningful, individuals should
claim basic functionality within their L2). This target pop-
ulation was chosen for five reasons: (a) to be representa-
tive of bilingual and multilingual groups most often studied
in research settings; (b) to accommodate the widest and
most diverse sample of bilingual and multilingual speak-
ers (with respect to specific languages, language history,
andlanguageuse);(c)toallowthequestionnairetobe
completed independently and with minimal support while
still providing meaningful data; (d) to span the variables
documented as most relevant in surveys of practicing
speech-language pathologists (e.g., Kohnert, Kennedy,
Glaze, Kan, & Carney, 2003; Roseberry-McKibbin et al.,
2005); and (e) to accommodate a tra de-off between col-
lecting a wide range of relevant measures while at the
same time capturing sufficient detail about specific as-
pects of bilingual status.
The LEAP-Q was constructed to assess bilingual ex-
perience and proficiency profiles in first and second lan-
guages, irrespective of the specific languages involved.
Therefore, in the current studies, different languages
served as L1 and L2, with L1 status signifying that the
language had been learned first (but not that it was
dominant or that proficiency in it had been attained). Dif-
ferences between subgroups of bilinguals (e.g., English
Spanish vs. SpanishEnglish) were not of interest in this
research; instead, all data were analyzed in terms of
L1 vers us L2, t hereby increasing the st atistic al power
of various analyses and facilitating the questionnaires
generalizability to different populations of adult bilingual
speakers. In both studies, the questionnaire was devel-
oped, administered, and normed in English only. Specific
objectives of the project were to establish internal validity
of the LEAP-Q, to establish criterion-based validity of the
LEAP-Q, to establish predictive relationships between
self-reported measures and performance on standardized
language tests, and to establish predictive relationships
between language history and self-reported proficiency
levels.
Study 1: Establishing Internal
Validity of the LEAP-Q
Study 1 aimed to establish internal validity of the
questionnaire by analyzing responses of a diverse group
of 52 bilinguals using factor analysis and multiple re-
gression analyses. Factor analysis has been routinely
used as a statistical method to uncover groups of vari-
ables that share variance patterns and are likely to mea-
sure the same construct. It has been successfully used
in questionnaire and scale development, revealing con-
structs underlying cultural identification (e.g., Zea,
Asner-Self, Birman, & Bu ki, 2003), emotional intelligence
(e.g., Tapia, 2001), well-being of elderly bilinguals (e.g.,
Tran, 1994), and caring behaviors (e.g., Wu, Larrabee, &
Putman, 2006). Factor analysis can uncover reliable under-
lying constructs despite possible surface dissimilarities in
the data. Cross-linguistic studies have demonstrated that
translating a questionnaire into a different language did
not change the structure of factors, suggesting that the
underlying constructs were not dependent on test language
(e.g., Abdel-Khalek, Tomas-Sabado, & Gomez-Benito, 2004;
Ferrer et al., 2006). For example, Wiebe and Penley (2005)
showed that English scales translated into Spanish main-
tained the same factor structure. In Study 1, bilinguals
944 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
answers to LEAP-Q questions designed to measure a sin-
gle construct were predicted to cluster together during
factor analysis, to show distinct patterns for each language,
and to yield factors indicative of L1 and L2 proficiency . It
was also expected that bilinguals language history would
predict self-reported proficiency levels in L1 and L2.
Method
Participants. The questionnaire was administered to
52 multilingual individuals (M =27.29years,SD =5.92;
29 women, 23 men). Participants were recruited from the
Northwestern University campus and the greater Chi-
cago metropolitan area communities. Participants varied
in their education level from 2 years of college to a doc-
toral degree (M = 18.04 years of education, SE =2.62;
range = 1527 years). None reported a hearing, language,
or learning disability. Of the 52 participants, 11 spoke two
languages, 19 spoke three languages, 12 spoke four lan-
guages, and 10 spoke five languages. Across participants,
34 languages were represented: American Sign Language,
Belorussian, Bengali, Cantonese, Croatian, Czech, Dutch,
English, Filipino, French, German, Hebrew, Hindi,
Hungarian, Italian, Japanese, K orean, Mandarin,
Malayalam, Marathi, Norwegian, Polish, Portuguese,
Punjabi, Romanian, Russian, Spanish, Swahili, Tamil,
Taiwanese, Telugu, Thai, Ukrainian, and Welsh.
Participants self-reported language history and pro-
ficiency measures can be found in Table 1. L2 acquisition
ages ranged from 0 to 15 years, representing both simul-
taneous and sequential bilinguals. Participants reported
being exposed to L1 most in the context of family, followed
by friends, reading, radio, TV, and independent language
study. Participants reported being exposed to L2 most in
the context of reading, followed by friends, TV, radio,
family, and independent language study. When asked
to report how different factors contributed to language
learning, participants reported that learning L1 relied
most on family, followed by friends, reading, TV, radio,
and self-instruction, whereas learning L2 relied most on
reading, followed by friends, TV, radio, family, and self-
instruction (see Table 1). When asked to repo rt proficiency
in each language, partici pants reported highest proficiency
for understanding, followed by speaking, reading, and
writing for both L1 and L2.
Materials and procedure. The development of the
questionnaire followed the steps outlined in Question-
naires in Second Language Research (Dörnyei, 2003,
pp. 6669) and included compilation of an initial item
pool, discussion of questions, omission of jargon, and clar-
ification and simplification of instructions and questions.
First, an expert with experience in questionnaire develop-
ment was consulted (the experts primary area of exper-
tise was audiology), and questionnaires in communication
sciences and disordersrelated fields were reviewed (e.g.,
Demorest & Erdman, 1987; Kohnert et al., 2003; Kuk,
Tyler, Russel, & Jordan, 1990; Roseberry-McKibbin et al.,
2005). Second, questions were chosen and prepared on
the basis of assessment materials previously used in ex-
perimental studies conducted in our laboratory, other lab-
oratories, and as described in the literature (Chincotta &
Underwood, 1998; Flege et al., 1999, 2002; Jia et al., 2002;
Vaid & Menon, 2000). Metacognitive judgment questions,
such as evaluations of the relative contribution of differ-
ent exposure variables to learning a language (e.g., in-
teracting with family/friends, reading, etc.), also were
included. Third, the d raft version of the LEAP-Q was
piloted with 8 bilingual and multilingual participants
(Marian, Blumenfeld, & Kaushanskaya, 2003). All pilot
participants spoke two or more languages and were rep-
resentative of the LEAP-Q target population. Fourth, on
the basis of participants responses and feedback, the
LEAP-Q was revised for clarity and succinctness to ac-
commodate efficient self-reporting across different lan-
guages. Questions were considered on an item-by-item
basis to ensure that none yielded missing values, outliers,
or insufficient variability. Fifth, the resulting items were
piloted with members of the Bilingualism and Psycho-
linguistic Research Laboratory and with members of the
Communication Sciences and Disorders department at
Northwestern University. On the basis of the feedback
received, the LEAP-Q was revised and distributed to the
participants in Study 1.
Domains assessed by the LEAP-Q included acquisi-
tion history, contexts of acquisition, present language use,
language preference and proficiency ratings (across the
four domains of language use: speaking, understanding,
reading, and writing), and accent ratings. Some questions,
such as those inquiring about ages of L2 acquisition, were
applicable to all bilinguals tested; other questions, such
as those inquiring about L1 learning from tapes, applied
to a subgroup of bilinguals only (e.g., individuals in im-
migrant communities who learned their first language
incompletely from their families and attempted to
maintain and expand L1 proficiency by means of self-
instruction). Questions pertaining to each language were
designed to be identical, to accommodate variability in
histories of L1 learning, and to maintain maximal flex-
ibility of the questionnaire. Participants completed the
questionnaire independently in approximately 25 min.
Analyses. Factor analysis was conducted to compare
statistical clustering of questions with accepted dimen-
sions of bilingual status. Seventy-seven attributes were
entered into the principal components analysis, which
served as the extraction method, and a varimax rotation
method was applied. The statistical software was given a
maximum of 100 iterations to converge on a factor solu-
tion, and the rotation converged in 51 iterations. Patterns
of variables within a single construct were examined,
Marian et al.: Bilingual LEAP Questionnaire 945
commonalities underlying clusterings of variables within
a single factor were identified, and the factor name was
logically deduced. For factors that included variables
with both positive and negative loadings, positive load-
ings provided inclusionary criteria and described the
underlying construct reflected by the factor, whereas neg-
ative loadings provided exclusionary criteria and indi-
cated an inverse relationship to the construct reflected
by the factor. The names were intended to capture the
nature of the variables that clustered together and to
suggest underlying commonalities among them. This
procedure is the standard way for performing factor anal-
ysis on large data sets (e.g., Tapia, 2001; Tran, 1994; Wu
et al., 2006; Zea et al., 2003). In this study, although some
constructs were expected to appear a priori, others were
determined only after they emerged from analyses.
It should be noted that the factor analysis in this
study was not used for combining items on the question-
naire into a single representative score. Although the
LEAP-Q was constructed so that multiple questions mea-
sured a single domain of language proficiency (e.g., four
questions targeted age of acquisition, four questions tar-
geted proficiency), the questions were not intended to
yield multi-item scales. For example, although the ques-
tions that clustered together in the L2 Competence fac-
tor were representative of a single underlying construct,
combining these questions into a subscale of language
competence may not be effective at measuring L2
Table 1. Self-reported language history and proficiency for participants in Study 1.
Language history measures
L1 history L2 history
MSD Range MSD Range
Self-reported proficiency
a
Understanding 4.86 0.46 3.005.00 4.20 0.90 2.005.00
Speaking 4.65 0.60 3.005.00 4.02 0.99 1.005.00
Reading 4.25 1.52 0.005.00 4.20 1.11 0.005.00
Writing 4.08 1.48 0.005.00 3.76 1.39 0.005.00
Age milestones (years)
Started learning 0.33 0.68 0.003.00 6.82 4.49 0.0015.00
Attained fluency 3.61 2.59 0.0010.00 14.31 8.90 0.0038.00
Started reading 4.5 3.21 0.0019.00 8.15 3.23 2.0013.00
Attained reading fluency 8.07 2.94 3.0015.00 15.10 7.74 5.0043.00
Immersion duration (years)
In a country 18.27 10.49 0.0047.00 7.85 8.36 0.0028.58
In a family 19.89 10.20 0.0047.00 4.54 7.98 0.0026.42
In a school 10.39 8.84 0.0030.00 5.57 7.34 0.0028.50
Contribution to language learning
b
From family 4.58 0.92 1.005.00 2.34 1.57 0.005.00
From friends 3.96 1.09 1.005.00 3.73 1.27 1.005.00
From reading 3.73 1.28 1.005.00 3.92 1.37 1.005.00
From TV 2.75 1.21 1.005.00 3.35 1.34 1.005.00
From radio 2.40 1.20 1.005.00 2.84 1.41 1.005.00
From self-instruction 1.43 1.04 0.005.00 1.92 1.21 0.005.00
Extent of language exposure
c
Family 4.32 1.04 1.005.00 2.00 1.30 1.005.00
Friends 3.59 1.30 1.005.00 3.35 1.28 1.005.00
Reading 3.00 1.47 1.005.00 3.67 1.53 1.005.00
TV 2.29 1.45 1.005.00 3.25 1.58 1.005.00
Radio 2.53 1.61 1.005.00 3.00 1.69 1.005.00
Independent study 1.08 0.35 1.003.00 1.24 0.66 1.004.00
Self-report of foreign accent
d
Perceived by self 0.56 1.10 0.005.00 1.87 1.49 0.005.00
Identified by others 0.38 0.79 0.004.00 2.19 1.42 0.004.00
Note. L1= first language; L2 = second language.
a
Range: 0 (none)to5(high).
b
Range: 0 (not a contributor)to5(most important contributor).
c
Range: 1 (not at all)to5(always).
d
Range: 0 (none)to5(very strong).
946 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
proficiency in a different sample of bilingual participants.
Theoretically, the construct of L2 Competence can be
expected to involve at least three different subdomains:
(a) age of acquisition, (b) length of immersion, and (c) self-
reported degree of language proficiency. Differences
across any of these domains would yield different bilin-
gual profiles. Moreover, differences in population statis-
tics would likely result in different factor structures
across studies. Given the complex and varied nature of L2
acquisition across populations, questions in the LEAP-Q
were intended to be considered separately and not to be
reduced to a limited number of subscales. Factor analysis
served as a tool for the determination of whether LEAP-Q
questions contributed to underlying constructs shaping
bilingual status and was, therefore, valid; it was not used
to define these constructs as universal scales.
In addition, multiple regression analyses were con-
ducted to examine the relationship between language his-
tory and language proficiency. Specifically, 16 attributes
of language history (i.e., acquisition and fluency ages,
learning environments, exposure variables) were entered
as independent variables into stepwise multiple regression
analyses, with self-reported proficiency in understanding,
speaking, reading, and writing as dependent variables.
Pearson R values, F and p values, and regression co-
efficients for the best predictor models are reported. (Note
that magnitudes of beta values in regression analyses
are, in part, dependent on the rating scale of the self-
reported or behavioral measure in question.) Regression
analyses in this study were used as exploratory tools
rather than for theory testing. Although some researchers
have found stepwise multiple regression analyses to be
controversial (e.g., Menard, 1995; Tabachnick & Fidell,
2007), these analyses were the most appropriate way of
examining the current data set because of the high num-
ber of independent variables (i.e., 16 language history
variables). Stepwise multiple regression analyses made it
possible to zero in on the variables that could account for
the most variability in self-ratings, which is an especially
useful approach when the number of independent vari-
ables is large (e.g., Mendenhall & Sincich, 1996). The pre-
dictive power of independent variables is marked by R
values, where R values of .5 or higher are considered
large, R values between .3 and .5 are considered mod-
erate, and R values between .1 and .3 are considered small
(Cohen, 1988).
Results
Sixteen factors with eigenvalues greater than 1 were
extracted from the data set by means of factor analysis.
Of these factors, the first 8 had eigenvalues greater than 3
and accounted for 76% of all variance (see Table 2). These
8 factors were assigned construct names indicative of
their components and are listed in order of variance ac-
counted for. Cronbachs alphas were calculated for each
factor to assess consistency of components within each
and to assess the extent to which questions captured
similar information. (Cronbachs alphas should be .7 or
higher for items in a set to be considered internally con-
sistent with each other, although cutoffs ranging from .60
to .80 have been used; see Miller, 1995.)
The first factor, accounting for the most variance,
included self- reported proficiency and comfo rt with speak-
ing, understanding, reading, and writing in L1; identifi-
cation with L1-associated culture; and preference for
reading in L1 (all positive loadings), as well as L1 accent
(negative loading; Cronbachs a = .85). The positive load-
ings of L1 proficiency variables together with the negative
loadings of L1 accent ratings suggested that this factor
was an index of cross-modal L1 Competence.
The second factor (in order of variance accounted for)
included age of initial L2 acquisition and age of attained
L2 fluency (positive loadings), and comfort and proficien-
cy in understanding L2 (negative loadings). Cronbachs
alpha could not be calculated for this factor because of
negative average covariance among items, which violated
reliability model assumptions. The negative covariance
resulted from the inverse relationship between age of
acquisition and language competencethat is, later
acquisition was associated with lower competence. Pos-
itive loadings for ages of acquisition suggested late
language development, whereas negative loadings for
proficiency variables suggested incomplete acquisition of
the second language. Together, the clustering of these
variables suggested that oral comprehension presented
a special challenge for late learners, and this factor was
interpreted as a measure of Late L2 Learning.
The third factor included total time exposed to L2;
exposure to L2 via TV, friends, radio, family, reading, and
classroom; proficiency and comfort with writing in L2;
and a preference for speaking L2 (all positive loadings)
as well as learning L2 from reading (negative loading;
Cronbachs a = .92). Positive loadings for L2 immersion
and proficiency variables indicated a common underlying
competence factor, whereas a negative loading for learn-
ing L2 from reading suggested that, in these bilinguals,
competence in L2 was related to social L2 immersion.
Together, these patterns were interpreted as an index
of L2 Competence.
The fourth factor included total time exposed to L1;
exposure to L1 via classroom, TV, radio, reading, and
friends (positive loadings); and learning L2 from reading
(negative loading; Cronbachs a = .80). The presence of an
L2 learning variable localized the underlying phenome-
non to an L2 context, whereas the positive loadings of L1
exposure variables suggested continued immersion in L1.
Therefore, this clustering was interpreted as a measure
of continued L1 exposurethat is, L1 Maintenance.
Marian et al.: Bilingual LEAP Questionnaire 947
Table 2. Factors yielded in Study 1.
Factor 1:
L1 Competence
Loading
values
Factor 2:
Late L2 Learning
Loading
values
Factor 3:
L2 Competence
Loading
values
Factor 4:
L1 Maintenance
Loading
values
Proficiency reading .947 Age became fluent .864 Exposure (% time) .923 L1 exposure to classes .916
Comfort understanding .910 Age began acquiring .859 Exposure to TV .908 L1 exposure to TV .914
Proficiency understanding .910 Age became fluent reader .855 Exposure to friends .861 L1 exposure to radio .831
Comfort writing .903 Comfort understanding .803 Exposure to radio .772 L1 exposure to reading .776
Proficiency writing .896 Age began reading .751 Writing proficiency .660 L2 learning from reading .727
Comfort reading .884 Proficiency understanding .697 Exposure to family .621 L1 exposure (% time) .627
Identified accent .788 Years in a country .681 Comfort writing .592 L1 exposure to friends .530
Comfort speaking .748 Learning from tapes .601 Preference to speak .590
Proficiency speaking .704 Proficiency speaking .580 Exposure to reading .564
Cultural identification .526 Exposure to classes .543
Perceived accent .517 Learning from reading .519
Preference to read .457
% variance 23.480 13.383 9.625 7.534
Cumulative variance 23.480 36.862 46.488 54.021
Factor 5:
Late L2 Immersion
Loading
values
Factor 6:
Media-Based Learning
Loading
values
Factor 7:
Non-Native Status
Loading
values
Factor 8:
Balanced Immersion
Loading
values
L1 years of class learning .728 L1 learning from TV .866 L2 perceived accent .839 L1 learning from friends .813
L2 years in workplace .725 L2 learning from TV .838 L2 identified accent .615 L2 years of schooling .627
L1 years in workplace .714 L1 learning from the radio .741 L2 cultural identification .602 L1 years in a family .622
Proficiency reading L2 .687 L2 learning from the radio .652 L1 age became fluent .590 L2 years in a classroom .541
L2 learning from friends .683 L2 comfort reading .476 L2 learning from family .519 L1 years in a country .499
L2 learning in a classroom .556
L1 years in school .476
% variance 6.424 6.226 5.049 4.232
Cumulative variance 60.445 66.671 71.720 75.952
948 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
The fifth factor included years exposed to L1 in a
language classroom, workplace, and general school set-
ting, and years exposed to L2 in the workplace (all posi-
tive loadings), as well as proficiency in reading L2 and
learning L2 from friends and in a foreign language
classroom (negative loadings; Cronbachs a = .30; how-
ever, note that Cronbachs alphas are likely to be under-
estimated in factors where both positive and negative
loadings are present). The positive loading of L2 work-
place immersion suggested adult L2 acquisition, whereas
positive loadings of a range of L1 immersion variables
(including workplace and school settings) suggested late
immigration from the L1-speaking country. This cluster-
ing of variables is likely to describe a subset of bilinguals
consisting of adult immigrants, who spent their formative
years in an L1-speaking environment and who were im-
mersed in L2 later in life. To account for the presence of
both L1 and L2 variables within this cluster, this factor
was interpreted to index Late L2 Immersion.
The sixth factor included learning L1 and L2 from
radio and TV (positive loadings), as well as comfort with
reading in L2 (negative loading; Cronbachs a =.75).
Positive loadings for both L1 and L2 suggested that the
construct underlying this factor was not specific to just
one language; instead, it was more likely to capture a
general trend of language learning from the media. This
pattern of variable loadings indicated a measure of lan-
guage learning within a popular culture framework and
was interpreted as a measure of Media-Based Learning.
The seventh factor included L2 accent as perceived
by the participant and as identified by others, age of at-
tained L1 fluency (positive loadings), and identification
with L2 culture and learning L2 from family (negative
loadings; Cr onbachs a = .24). Positive loadings for acqui-
sition ages and for accent judgments indicated late L2
acquisition, and negative loadings for identification with
L2 culture and learning L2 from family indicated lack of
assimilation into the L2 environment. Age of attained L1
fluency loaded positively, suggesting that this cluster was
specific to a subgroup of bilinguals who experienced a
discontinuity in use of their L1. Together, these patterns
were interpreted to index Non-Native Status.
The eighth factor included years spent in an L1 fam-
ily and country, years exposed to L2 in a language class-
room and in a general school setting (all positive loadings),
and learning L1 from friends (negative loading; Cronbachs
a = .27). Positive loadings from both L1 and L2 variables
suggested that the underlying phenomenon was common
to both languages, whereas the negative loading for learn-
ing L1 from friends suggested relatively early immi-
gration from the L1-speaking country. Together, these
patterns suggested a balanced bilingual profile, where
both L1 immersion and L2 immersion were important.
Therefore, this factor was probably descriptive of a subset
of bilinguals who were born in an L1 country but were
schooled mostly in an L2 country, and it was interpreted
to reflect Balanced Immersion.
The language history measures that predicted pro-
ficiency in understanding, speaking, reading, and writing
in L1 and L2 are reported in Table 3. The results of the
stepwise multiple regression analyses include regression
coefficients (marking the relative importance of each in-
dependent variable that entered the model) and statistics
describing the fit of the model. In addition to standard-
ized regression beta, variation inflation factors (VIFs) are
included to measure the correlation among independent
variables (lack of such correlation is a basic assump tion in
regression analysis). It is generally accepted that VIF val-
ues greater than 10 signal multicollinearity and singu-
larity problems (Mendenhall & Sincich, 1996). In this
study, VIF values of independent variables ranged from
1.0 to 1.4, suggesting that no multicollinearity/singula rity
problems were present.
Discussion
Participants in Study 1 reported high levels of pro-
ficiency and extensive immersion in both languages. Across
modalities, L1 proficiency was higher than L2 proficiency,
with the largest difference in understanding and the
smallest difference in reading. It is possible that the em-
phasis on reading in this study is reflective of the specific
sample of bilinguals in Study 1, who were recruited pri-
marily from academic communities. In order to succeed in
an L2 academic environment, reading in L2 was likely to
be particularly important. Therefore, bilinguals in this
study may have been more likely to judge their overall L2
competence on the basis of L2 reading skills versus other
skills. The emphasis on L2 reading evident in proficiency
ratings was echoed in participants reports that reading
experiences (e.g., learning via reading and exposure to
reading) contributed most to L2 competence, whereas
family-based experiences (e.g., learning language from
family and exposure to family) contributed most to L1
competence. These patterns of reading-oriented L2 acqui-
sition accurately reflect typical language learning pat-
terns for sequential bilinguals, where a native language
is acquired within a family environment and a second
language is often acquired on entrance into schooling and
takes place in a classroom environment involving explicit
reading instructions.
Factor analysis yielded component groupings that
accounted for most of the variance in bilinguals self-
reported data, suggesting that questions on the LEAP-Q
were broad enough to capture variability in the bilingual
population that was sampled in this study. The clusters
that emerged reflected underlying dimensions of bilingual-
ism, with questions that clustered together measuring the
same construct. The first four factors (L1 Competence,
Marian et al.: Bilingual LEAP Questionnaire 949
Late L2 Learning, L2 Competence, and L1 Maintenance)
were general in nature, provided information shared by
all bilinguals, and accounted for more than half of the
variance in the data. The remaining factors (Late L2
Immersion, Media-Based Learning, Non-Native Status,
and Balanced Immersion) grew increasingly specific and
appeared to be driven by bilingual subgroups. In addition,
the measures of variance in the data set accounted for
by each factor (indicated by eigenvalues) were augmented
by measures of consistency within each observed factor
(indicated by Cronbachs alphas). The four factors with
highest Cronbachs alphas were general in nature: L1
Competence, L2 Competence, L1 Maintenance, and
Media-Based Learning. Conversely, the three factors
with lower Cronbachs alphasLate L2 Immersion, Non-
Native Status, and Balanced Immersionmight have
emerged from characteristics of specific subgroups in the
sample tested. Together, the eigenvalues and Cronbachs
alphas reveal that L1 Competence, L2 Competence, and
L1 Maintenance accounted for much of variance in the
data and w ere highly consistent internally. ( Late L2
Learning accounted for approximately 13% of the vari-
ance but did not lend itself to statistical evaluation of in-
ternal consistency because of negative covariance across
age and ability components. Finally, Media-Based Learn-
ing had high internal consistency but accounted for rela-
tively little variance in the data, about 6%.)
The structure of the L1 Competence and L2 Compe-
tence factors suggests that the constructs of proficiency
in the first and second language share subcomponents.
However, some differences in factorial structures were
also observed. For instance, unlike the L1 Competence
factor, the L2 Competence factor included a large number
of age-of-acquisition variables. The significant contribu-
tion of maturation variables to competence in L2 is con-
sistent with previous studies that have identified L2
acquisition age as an important determiner of L2 com-
petence (e.g., Flege et al., 1999, 2002; Jia et al., 2 002;
Johnson & Newport, 1989; Vaid & Menon, 2000). More-
over, the structure of the L2 Competence factor con-
firms Hyltenstam and Abrahamssons (2003) model of
L2 acquisition, which postulates interplay between age
of acqui sition and environmental variables in shaping
L2 attainment.
Although some factors may include variables that
appear similar, they represent constructs that differ in
nuanced ways. For example, L2 learning in a classroom
loads negatively onto Factor 5 ( Late L2 Immersion), and
years in an L2 classroom loads positively onto Factor 8
(Balanced Immersion). The two questions may contribute
to overall language proficiency in distinct ways. Classroom
experience as a contributor to learning L2 is a subjective,
metacognitive self-assessment measure, whereas dura-
tion of L2 classroom exposure is an objective temporal
Table 3. Multiple regression analyses for Study 1: Language history predictors of proficiency, including regression coefficients (B and b) and
corresponding variance inflation factors (VIFs), as well as evaluators of the fit of the model (R, R
2
, and Fs).
Predictee, F test Predictor
Regression coefficients Fit of model
B SE of B b VIF RR
2
Understanding L1 Years in an L1 country 0.02 0.01 .44 1.09 .55 .31
F(2, 50) = 12.7, p < .001 Learning L1 from reading 0.14 0.05 .40 1.09 .67 .45
Speaking L1 Years in an L1 country 0.04 0.01 .55 1.04 .61 .38
F(2, 50) = 15.1, p < .001 Exposure to L1 friends 0.17 0.06 .35 1.04 .70 .49
Reading L1 Years in an L1 country 0.06 0.02 .41 1.09 .52 .27
F(2, 50) = 11.5, p < .001 Learning L1 from reading 0.45 0.16 .41 1.09 .65 .43
Writing L1 Years in an L1 country 0.11 0.02 .61 1.09 .74 .55
F(2, 50) = 29.4, p < .001 Learning L1 from reading 0.46 0.13 .40 1.09 .83 .69
Understanding L2 Age began reading in L2 0.12 0.03 .52 1.00 .52 .27
F(1, 50) = 11.6, p < .01
Speaking L2 L2 exposure in family 0.35 0.09 .57 1.33 .48 .23
F(4, 50) = 7.4, p < .001 Age began reading in L2 0.12 0.04 .45 1.20 .60 .35
L2 classroom learning 0.28 0.10 .42 1.32 .66 .43
L2 exposure in classroom 0.14 0.07 .32 1.43 .71 .51
Reading L2 Learning L2 from reading 0.35 0.10 .50 1.07 .51 .26
F(3, 50) = 9.8, p<.001 Learning L2 from friends 0.34 0.11 .44 1.13 .60 .36
Learning L2 from radio 0.24 0.08 .39 1.13 .70 .50
Writing L2 Exposure to L2 friends 0.52 0.11 .64 1.05 .58 .33
F(2, 50) = 11.5, p<.001 Years in an L2 workplace 0.06 0.03 .31 1.05 .65 .43
950 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
measure. Contribution to learning may reflect learning
style, whereas years of exposure measures duration of
experience (amount of time spent in a classroom is not
always indicative of language proficiency). This distinc-
tion is reflected in differential loadings of these two vari-
ables onto Factors 5 and 8.
Multiple regression analyses were used to generate
predictive equations for self-reported proficiency levels in
L1 and L2. The results suggested that different experien-
tial variables predicted proficiency in the two languages.
For instance, L1 proficiency levels across modalities were
consistently predicted by years spent in an L1 country,
whereas L2 proficiency levels were predicted by L2
acquisition ages. In contrast to L1 regression models,
years of immersion (in this case, length of time spent in an
L2 workplace) predicted only proficiency in writing L2,
but not proficiency in speaking, understanding, or read-
ing L2. Moreover, although similar experiential variables
predicted proficiency levels in L1 across understanding,
reading, speaking, and writing, predictors of proficiency
levels in L2 were more varied. For instance, L1 pro-
ficiency levels were consistently predicted by time spent
in an L1 country. Conversely, predictors of L2 proficiency
differed: For example, proficiency in understanding was
predicted primarily by age when participants began read-
ing L2, and proficiency in speaking was predicted
primarily by exposure to L2 in family and classroom
environments. The greater variability and larger number
of language history predictors for L2 is likely due to
different acquisition patterns for the two languages, with
L2 acquisition more varied across settings and modalities
relative to L1. Greater variability in L2 acquisition pat-
terns is probably due to highly diverse experiences asso-
ciated with learning a second language (relative to a
native language, the acquisition of which is less variable).
For example, whereas some participants learned L2 in an
immersion-type setting (because of immigration, or
studying and working abroad), others learned L2 in a
classroom environment.
The results of Study 1 prompted exclusion of four
measures from the questionnaire. First, measures of com-
fort across modalities (e.g., How comfortable are you
speaking, understanding, reading, and writing in a lan-
guage?) were excluded. These questions yielded val-
ues that were similar to those yielded by profic iency
measuresthat is, the two patterned identically, corre-
lated significantly, and were highly predictive of each
other. Specifically, proficiency measures were predictive
of comfort measures for speaking (L1: R =.9,F[1, 50] =
167, p <.001;L2:R =.9,F[1, 50] = 264, p < .001),
understanding (L1: R =.95,F[1, 50] = 468, p < .001; L2:
R =.9,F[1, 50] = 332, p < .001), reading (L1: R = .97,
F[1, 50] = 881, p < .001; L2: R =.9,F[1, 50] = 181, p < .001),
and writing (L1: R = .99, F[1, 50] = 1760, p <.001;
L2: R =.96,F[1, 50] = 561, p < .001). Second, writing
proficiency was excluded because of its close relationship
to reading proficiencyspecifically , proficiency in reading
L1 predicted proficiency in writing L1, R =.9,F(1, 50) =
365, p < .001, and proficiency in reading L2 predicted
proficiency in writing L2, R =.8,F(1, 50) = 108, p < .001.
The hi gh predictability of wr iting proficiency f rom
measures of reading proficiency suggested that a separate
question assessing writing proficiency would not pro-
vide additional new information. Third, current classroom
exposure was excluded because of its close relationship to
current reading exposure, with reading exposure in L1
predicting ratings of classroom exposure in L1, R =.7,
F(1, 50) = 34, p < .001, and reading exposure in L2 predict-
ing ratings of classroom exposure in L2, R =.6,F(1, 50) =
33, p < .001. Because classroom exposure was generally
not predictive of self-reported proficiency ratings, and
because the extent of classroom exposure could be reli-
ably deduced on the basis of reading exposure, it was
omitted from the final version of the LEAP-Q. Fourth,
percentage of bilingual contacts was excluded because it
did not correlate with any other measures and did not load
onto any factor in the factor analysis. Because the aim of
this study was to construct an internally consistent self-
assessment measure, a question that was not related to
any other questions in the questionnaire was deemed
unreliable and uninformative. As a result of these changes,
seven questions per language were omitted, shortening
questionnaire completion time without losing predictive
value. In addition, because of limited variability (15) in
questions that required responses on a scale, the ranges of
values on all scales were increased (110). Finally, the
questionna ire was transferred into digital format. Pull-
down menus were added to questions that required
responses on scales, thus facilitating questionnaire com-
pletion and subsequent data extraction fo r analyses. In
the digitized version, once the participant indicated a spe-
cific language, the language was filled in automatically
throughout the questionnaire. The revised version of the
questionnaire was used in Study 2 and required approx-
imately 15 min for bilinguals to complete.
Study 2: Establishing
Criterion-Based Validity
The objectives of Study 2 were to confirm the inter-
nal validity of the LEAP-Q in a more homogeneous sam-
ple of bilinguals through factor analysis as well as to
establish criterion-referenced validity by comparing self-
reported and standardized proficiency measures using
correlation and regression analyses. We made the
following five predictions:
1. Bilinguals answers to questions that referred to the
same underlying aspects of bilingual profiles would
Marian et al.: Bilingual LEAP Questionnaire 951
pattern together in factor analysis and would mirror
factors revealed in Study 1, therefore supporting the
internal validity of the questionnaire.
2. Similar to Study 1, bilinguals self-reported language
history would be predictive of their self-reported pro-
ficiency levels in L1 and L2.
3. Self-reported proficiency would correlate with and pre-
dict performance on standardized language measures.
4. In turn, performance on standardized language tests
would predict self-reported proficiency, providing in-
formation about the metaknowledge on which bilin-
guals rely when estimating their language proficiency.
5. Language history would predict performance on stan-
dardized language measures.
Method
Participants. Fifty bilingual speakers of English and
Spanish participated (M =26.7,SD = 10.4; 31 women,
19 men). Of these, 18 were SpanishEnglish bilinguals
(native speakers of Spanish who acquired English as a
second language), and 32 were EnglishSpanish bilin-
guals (native speakers of English who acquired Spanish
as a second language). Participants were recruited from
the Northwestern University and the Chicago metropol-
itan area communities by means of flyers. None reported
a hearing, language, or learning disability. L1 and L2
acquisition ages, language histories, and language per-
formance information are presented in Tables 4 and 5.
L2 acquisition ages ranged from 0 to 23 years of age,
Table 4. Self-reported language history and proficiency for participants in Study 2.
Language history measures
L1 history L2 history
MSD Range MSD Range
Self-reported proficiency
a
Understanding 9.58 0.93 6.0010.00 7.92 2.33 2.0010.00
Speaking 9.32 1.15 6.0010.00 7.74 2.05 2.0010.00
Reading 9.26 1.26 5.0010.00 8.02 1.97 2.0010.00
Age milestones (years)
Started learning 1.08 1.75 0.0011.00 8.25 5.95 0.0023.00
Attained fluency 4.00 2.81 0.0013.00 15.10 7.44 2.0039.00
Started reading 5.5 2.24 3.0014.00 11.52 5.44 3.0025.00
Became fluent reading 7.92 3.27 4.0041.00 15.09 7.44 2.0039.00
Immersion duration (years)
Country 17.6 10.5 0.0054.00 11.3 12.2 0.0044.00
Family 22.65 7.96 0.0055.00 8.72 12.18 0.0037.00
School 17.62 8.46 3.0049.00 9.55 10.66 0.0049.00
Contribution to language learning
b
From family 8.94 1.97 1.0010.00 3.58 3.76 0.0010.00
From friends 7.3 2.74 0.0010.00 6.26 3.12 0.0010.00
From reading 7.54 2.69 0.0010.00 7.32 2.03 3.0010.00
From TV 4.88 2.85 0.0010.00 5.08 3.11 0.0010.00
From radio 3.14 2.84 0.0010.00 3.64 3.39 0.0010.00
From self-instruction 1.16 2.58 0.0010.00 3.32 3.63 0.0010.00
Extent of language exposure
c
To family 7.62 3.02 0.0010.00 3.16 3.46 0.0010.00
To friends 7.34 3.34 0.0010.00 5.18 3.81 0.0010.00
To reading 6.26 3.33 0.0010.00 5.56 3.43 0.0010.00
To TV 6.16 3.47 0.0010.00 4.6 3.87 0.0010.00
To radio 6.48 3.23 0.0010.00 4.9 3.70 0.0010.00
Self-instruction 1.32 2.89 0.0010.00 2.18 3.02 0.0010.00
Self-reported foreign accent
d
Perceived by self 0.70 1.25 0.006.00 3.28 2.62 0.009.00
Identified by others 0.82 1.85 0.008.00 4.06 3.38 0.0010.00
a
Range: 0 (none)to10(perfect).
b
Range: 0 (not a contributor)to10(most important contributor).
c
Range: 0 (never)to10(always).
d
Range: 0 (none)to10(pervasive).
952 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
representing both simultaneous and sequential bilinguals.
Participants varied in their education levels from high
school to graduate school (M = 16 years of education,
SE =2.5,range=1122 years; note that in some countries,
a high school education is equivalent to fewer than
12 years). Participants reported being exposed to L1 most
in the context of family, followed by friends, radio, read-
ing, TV, and self-instruction. They reported being exposed
to L2 most in the context of reading, followed by friends,
radio, TV, family, and self-instruction. When asked to re-
port how different factors contributed to language learn-
ing, participants reported that learning L1 relied most on
family, followed by reading, friends, TV, radio, and self-
instruction, and they reported that learning L2 relied
most on reading, followed by friends, TV, radio, family,
and self-instruction. Participants performed better in L1
than in L2 across a range of behavioral measures (see
Table 5 for means and statistical comparisons).
Materials and procedure. The revised version of the
LEAP-Q was administered to all participants at the start
of the experimental session on a computer. Participants
independently completed the LEAP-Q in English. On
completion, participants were administered a battery of
standardized behavioral measures of language ability.
These included subtests from the WoodcockJohnson
Tests of Achievement (Woodcock, McGrew, & Mather,
2001), the WoodcockMuñoz Tests of Achievement
(Muñoz-Sandoval, Woodcock, McGrew, & Mather, 2005),
and the Peabody Picture Vocabulary Test in English
(PPVT, Dunn & Dunn, 1997) and Spanish (Test de
Vocabulario en Imágenes Peabody [TVIP]; Dunn, Padilla,
Lugo, & Dunn, 1986). In addition, sentence grammati-
cality judgment tasks were constructed on the basis of
previous materials (Bedoya et al., 2005; DeKeyser, 2000;
Johnson & Newport, 1989). All measures were adminis-
tered in language blocks, with half the participants re-
ceiving the Spanish measures first and half receiving the
English measures first. Test administrators were highly
proficient in both languages. Specifically, the following
seven behavioral measures were administered:
1. A reading fluency test (Subtest 2 of the Woodcock
Johnson T est of Achievement in English and its equiv-
alent WoodcockMuñoz version in Spanish). This test
required participants to read as many sentences as
possible within a 3-min interval and to decide whether
each sentence was true or false.
2. A passage comprehension test (Subtest 9 of the
WoodcockJohnson Test of Achievement in English
and its equivalent WoodcockMuñoz version i n
Spanish). This test required participants to read
passages and supply missing words.
3. A productive picture vocabulary test (Subtest 14 of the
WoodcockJohnson Test of Achievement in English
and its equivalent WoodcockMuñoz version in Span-
ish). This test required participants to name pictures.
4. An oral comprehension test (Subtest 15 of the
WoodcockJohnson Test of Achievement in English
and its equivalent WoodcockMuñoz version i n
Spanish). This test required participants to listen to
passages and supply missing words.
5. A sound awareness test (Subtest 21 of the Woodcock
Johnson Test of Achievement in English and its
equivalent WoodcockMuñoz version in Spanish).
This test required participants to complete a rhym-
ing task, a sound deletion task, a sound substitution
task, and a sound reversal task.
Table 5. Standardized proficiency measures for Study 2.
Measure
L1 performance L2 performance L1 L2 comparisons
MSD
Performance
range MSD
Performance
range t tests and effect sizes (partial h
2
)
WoodcockJohnson/WoodcockMuñoz
Reading Fluency (percentile) 62 28 1299.9 35 27 297 t(49) = 4.56, p<.01, h
2
= .30
Oral Comprehension (percentile) 64 21 398 36 30 0.197 t(49) = 4.78, p<.01, h
2
= .33
Passage Comprehension (percentile) 59 26 596 38 31 0.193 t(49) = 3.32, p<.01, h
2
= .18
Productive Vocabulary (percentile) 47 28 0.199.9 20 23 0.175 t (49) = 4.51, p<.01, h
2
= .29
Sound Awareness (percentile) 50 20 1995 35 22 6 89 t(49) = 3.74, p < .01, h
2
= .26
PPVT/TVIP measures (percentile) t(49) = 4.76, p<.01, h
2
= .32
80 16 4299.9 60 27 0.196
Grammaticality judgments
Accuracies (percentage) 80 20 38100 70 20 3196 t(49) = 3.26, p<.01, h
2
= .18
Response latencies (ms) 2,786 1,033 9805,290 3,842 1,445 1,7819,861 t(49) = 4.01, p<.01, h
2
= .25
Note. PPVT = Peabody Picture Vocabulary Test; TVIP = Test de Vocabulario en Imágenes Peabody.
Marian et al.: Bilingual LEAP Questionnaire 953
6. A receptive vocabulary test (PPVT/TVIP). This test
required participants to identify pictures in response
to auditory instructions.
7. A grammaticality judgment test ( based on Bedoya
et al., 2005; DeKeyser, 2000; Johnson & Newport,
1989). This test required participants to read 50
sentences in English and 50 sentences in Spanish
and to judge whether these sentences were gram-
matically correct. For each sentence, a grammati-
cally correct and a grammatically incorrect version
were presented at least 20 trials apart from each
other. The test was administered on a computer
screen, and participants identified sentences as cor-
rect or incorrect by pressing keys on the keyboard.
Sentences were matched in length, both in terms of
number of letters ( because the test was adminis-
tered visually) and in terms of number of phonemes
(in case subvocal articulation strategies were used).
The number of letters in English (M = 30.0, SD =
7.3) and Spanish sentences (M = 28.7, SD = 7.3) did
not differ, t(98) = 0.9, p > .1; the number of pho-
nemes in English (M = 24.7, SD = 5.7) and Spanish
sentences (M = 27.0, SD = 7.0) also did not differ,
t(98) = 1.8, p > .05.
Data Coding and Analyses
For all standardized measures, participant scores
were coded in terms of percentile rankings on a scale from
0 to 100. Percentile rankings were derived following the
normative data available for each of the standardized
tests (PPVT/TVIP [see Dunn & Dunn, 1997, and Dunn
et al., 1986] and the WoodcockJohnson Tests of Achieve-
ment and WoodcockMuñoz Tests of Achievement [see
Woodcock et al., 2001, and Muñoz-Sandoval et al., 2005]).
Grammaticality judgment accuracy was coded as propor-
tion correct, and grammaticality judgment latencies were
coded in milliseconds. Grammaticality judgment laten-
cies were measured as the duration of time between the
onset of sentence presentation and the participantsre-
sponse. Only the latencies for correctly identified sen-
tences were included in the latency analyses. Outliers
(latencies longer or shorter than 3 SDs from the partic-
ipants mean) were excluded from analyses.
Factor analysis, as well as correlation and multiple
regression analyses, were conducted. Multiple regression
analyses were run to examine predictors of proficiency
levels in bilinguals. First, 16 attributes of language his-
tory were entered as independent variables into stepwise
multiple regression analyses, and self-reported proficiency
in understanding, speaking, and reading were entered
as dependent variables. Second, questionnaire-based lan-
guage history attributes were entered as independent
variables into stepwise multiple regression analyses, and
behaviorally established proficiency measures were en-
tered as dependent variables. Third, results of standard-
ized language tests were entered as indep endent
variables into stepwise multiple regression analyses, and
self-reported proficiencies in understanding, speaking,
and reading were entered as dependent variables.
Results
Fourteen factors with eigenvalues greater than 1
were extracted from the data set using factor analysis. Of
these factors, the first 5 had eigenvalues greater than 3.
For consistency across the two studies, the first 8 factors
were assigned construct names indicative of their com-
ponents and are listed in order of variance accounted for.
Each of these factors had eigenvalues greater than 2.2,
and together they accounted for 73.5% of variance in the
data (see Table 6). Cronbachs alpha was calculated for
each factor separately, yielding values ranging from .31 to
.92, suggesting overall consistency of components within
each factor.
The first factor (accounting for the most variance)
included total time exposed to L2; exposure to L2 via TV,
reading, friends, and radio; years spent in an L2 country;
proficiency reading and speaking L2; and preference to
read in L2 (all positive loadings), as well as total time
exposed to L1; exposure to L1 via reading, TV, radio, and
friends; preference to read in L1; learning L1 from read-
ing; and accent in L2 as identified by others (all negative
loadings). The observed distinct patterns for L2 versus L1
suggested separate subcomponents (inversely related) for
each language within a single factor. The finding that in
this factor, L2-related variables loaded positively and L1-
related variables loaded negatively may indicate that this
group of bilinguals was L2 dominant. Moreover, the in-
verse relationship between L1 and L2 variables is con-
sistent with other studies of relative L1/L2 language
competence (e.g., Flege et al., 2002; Harley et al., 1990).
To circumvent a violation of assumptions for obtaining
reliability values (i.e., negative covariance between L2
and L1 components), Cronbachsalphawascomputedsep-
arately for L2 (.88) and L1 (.92) components. This factor
was taken to index Relative L2L1 Competence.
The second factor (in order of variance accounted for)
included age that participant began learning to read in
L1; age of becoming a fluent L1 reader; and learning L1
from tapes (all positive loadings); as well as proficiency in
speaking, reading, and understanding L1 (negative load-
ings; Cronbachs a = .31). Positive loadings of acquisition
age variables suggested a late-acquisition profile, whereas
negative loadings of proficiency variables suggested an
incompletely acquired L1. This L1 learning profile may
be characteristic of bilinguals who immigrated from an
L1-speaking country early in life and/or of bilinguals who
954 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
Table 6. Factors yielded in Study 2.
Factor 1:
Relative L2L1 Competence
Loading
values
Factor 2:
L1 Learning
Loading
values
Factor 3:
Late L2 Learning
Loading
values
Factor 4:
L1 Nondominant Status
Loading
values
L1 exposure to reading .872 Speaking proficiency .929 L2 learning from the radio .755 L1 age began acquiring .871
L2 exposure to TV .866 Age began reading .824 L2 age when fluent reader .724 L1 learned from family .823
L2 exposure to reading .844 Reading proficiency .747 L2 age when fluent .654 L1 age when fluent .624
L1 exposure to TV .811 Age when fluent reader .745 L2 age began reading .605 L2 preference to speak .577
L2 exposure to friends .804 Comprehension proficiency .709 L2 exposure to self-instruction .561 L1 identified accent .526
L2 exposure to radio .777 Learning from tapes .629 L2 learning from tapes .557 L1 exposure to family .468
L1 exposure to radio .773 L1 perceived accent .548
L1 exposure (% time) .771 L2 perceived accent .452
L1 exposure to friends .762
L2 exposure (% time) .759
L2 reading proficiency .754
L1 preference to read .713
L2 preference to read .696
L2 speaking proficiency .580
L1 learning from reading .522
L2 identified accent .496
L2 years in the country .474
% variance 25.296 12.425 9.615 7.471
Cumulative variance 25.296 37.722 47.337 54.807
Factor 5:
L2 Immersion
Loading
values
Factor 6:
L1 Immersion
Loading
values
Factor 7:
L2 Nonacculturation
Loading
values
Factor 8:
Media-Based L1 Learning
Loading
values
Exposure to family .898 L1 years of schooling .885 L2 age when began acquiring 0.854 L1 Learning from radio .845
Years in family .894 L1 years in family .818 L2 cultural identification 0.713 L1 Learning from TV .816
Learning from family .747 Chronological age .766
Years of schooling .537 L1 years in country .740
L2 learning from TV .503
% variance 5.933 4.860 4.095 3.835
Cumulative variance 60.740 65.600 69.695 73.530
Marian et al.: Bilingual LEAP Questionnaire 955
learned L1 at home but lived and received schooling from
an early age in an L2 environment and made an active
effort to maintain L1. These patterns were interpreted as
reflecting L1 Learning.
The third factor included learning L2 from radio
and language tapes, exposure to L2 through independent
study, ages of becoming a fluent L2 speaker and reader,
and self-perceived accent in L2 (all positive loadings), as
well as self-perceived accent in L1 (negative loading;
Cronbachs a = .77). Positive loadings of acquisition age
and self-instruction variables suggest late and incom-
plete acquisition of the second language, whereas positive
loadings of L2 accent and negative loadings of L1 accent
suggest higher L1 fluency than L2 fluency . Together , these
patterns were interpreted as indexing Late L2 Learning.
Thefourthfactorincludedagesof L1acquisition,ages
of attained L1 fluency, L1 accent as identified by others,
and preference to speak L2 (all positive loadings), as well
as exposure to and learning from an L1-speaking fam-
ily (negative loadings), suggesting limited L1 exposure.
Cronbachs alpha could not be calculated for this factor
because of the negative average covariance among items,
a violation of reliability model assumptions. Positive load-
ings of acquisition age and accent variables indicate lack
of fluency in L1, whereas negative loadings of family ex-
posure in L1 suggest lack of immersion in an L1 environ-
ment. These variables are likely to describe a subset of
bilinguals for whom L1 is no longer a dominant language
and who prefer to use L2 in daily life. Therefore, this fac-
tor was interpreted as indexing L1 Nondominant Status.
The fifth factor included L2 family-based components
(e.g., current exposure to, years spent in, and learning
from an L2-speaking family) as well as years of L2
schooling (all positive loadings; Cronbachs a = .86). This
factor was interpreted to suggest an overall measure of
interactive L2 Immersion.
The sixth factor included years spent in an L1-
speaking school, family, and country; participants chro-
nological age (all positive loadings); and learning L2 from
TV (negative loading: Cronbachs a = .50). The negative
loading of L2 learning variables may be indicative of a
subset of bilinguals who maintained their first language
and had minimal exposure to their second language. This
factor was interpreted as an overall measure of interac-
tive L1 Immersion.
The seventh factor included age of L2 acquisition
(positive loading) as well as identification with L2 culture
(negative loading). Cronbachs alpha could not be calcu-
lated for this factor because of the negative average
covariance among items, a violation of reliability model
assumptions. The positive loading of L2 acquisition age
onto this factor indicated a late L2 learning profile,
whereas the negative loading of L2 cultural identification
suggested lack of acculturation within the L2-speaking
country. Therefore, this factor was interpreted as index-
ing L2 Nonacculturation.
The eighth factor included learning L1 from radio
and TV (positive loadings, Cronbachs a = .77) and was
interpreted to reflect Media-Based L1 Learning.
Establishing predictive relationships using multiple
regressions. Regression analyses are reported in Table 7,
together with regression coefficients (marking the rela-
tive importance of each independent va riable that
entered the model) and statistics describing the fit of
the model (B and b coefficients, VIF values, R and R
2
values, and F tests). VIFs associated with each indepen-
dent variable ranged from 1.00 to 2.46, suggesting that no
multicollinearity or singularity problems were present.
The language history measures that predicted
self-reported proficiency in understanding, speaking,
and reading are reported in the top panel of Table 7, self-
reported proficiency measures that predicted behavioral
performance are reported in the second panel of Table 7,
language history measures that predicted behavioral
proficiency are reported in the third panel of Table 7,
and behavioral measures that predicted self-reported
proficiency are reported in the bottom panel of Table 7.
Correlations between behavioral and self-reported
measures. To establish criterion-based validity of self-
reported proficiency measures, Pearsons R correlation
analyses between self-reported and behavioral proficien-
cy measures were conducted within each language and
processing modality (see Table 8). The results yielded
strong positive correlations between standardized behav-
ioral measures (i.e., reading fluency, passage comprehen-
sion, productive vocabulary, oral comprehension, and
grammaticality judgments) and self-reported measures
of understanding, speaking, and reading L1 and L2. Al-
though performance on standardized measures of sound
awareness and receptive vocabulary did not relate to self-
reported L1 proficiency, it was significantly related to
self-reported L2 proficiency. The majority of standardized
measures correlated more strongly with self-reported L2
proficiency than with self-reported L1 proficiency (with
the exception of grammaticality judgment latencies,
which correlated stronger with self-reported L1 proficiency).
For L1, self-reported proficiency measures correlated most
strongly with standardized behavioral measures of oral
comprehension. For L2, the highest correlation val-
ues were obtained for passage comprehension and oral
comprehension.
Discussion
The results of Study 2 confirmed questionnaire-
based predictors of self-reported proficiency, identified
questionnaire-based predictors of behavioral language
performance, and revealed behavioral predictors of
956 Journal of Speech, Language, and Hearing Research Vol. 50 940967 August 2007
Table 7. Multiple regression analyses for Study 2: Language history predictors of self -reported proficiency, self-reported proficiency predictors of
behavioral performance, language history predictors of behavioral performance, and behavioral predictors of self-reported proficiency.
Predictee, F test Predictor
Regression coefficients Fit of model
B SE of B b VIF RR
2
Language history predictors of
self-reported proficiency
Comprehending L1 Age began reading L1 0.20 0.06 .37 1.00 .37 .13
F(1, 50) = 7.10, p<.05
Speaking L1 Age began reading L1 0.28 0.06 .54 1.05 .39 .39
F(2, 50) = 24.58, p<.001 0.13 0.04 .38 1.05 .72 .52
Reading L1 Exposure to L1 reading 0.18 0.05 .48 1.00 .48 .23
F(1, 50) = 13.57, p < .001
Comprehending L2 Exposure to L2 friends 0.44 0.07 .68 1.01 .72 .52
F(3, 50) = 21.6, p<.001 Learning from L2 family 0.20 0.07 .24 1.10 .78 .61
Age began acquiring L2 0.10 0.05 .22 1.11 .81 .52
Speaking L2 Exposure to L2 friends 0.39 0.06 .71 1.00 .72 .52
F(2, 50) = 29.4, p<.001 Learning from L2 family 0.17 0.06 .31 1.00 .79 .62
Reading L2 Exposure to L2 reading 0.28 0.09 .50 2.15 .74 .54
F(2, 50) = 26.02, p<.001 Exposure to L2 friends 0.16 0.08 .33 2.15 .77 .59
Self-reported proficiency predictors of
be