Metodološkizvezki,Vol. 14,No. 1,2017,1–17
SurveyModeasaModeratorofContextEffects
MayDoušak
1
Abstract
After decades of survey quality research and many experiments on the mode of
data collection as a source of data inequivalence, there is still no exact explanation
of how the mixed-mode design impacts the collected data. While not providing a
definitive answer, we propose a new explanation of the mode effect and how it cor-
relates with other sources of errors in surveys. By reviewing the relevant literature,
we develop theoretical arguments indicating that the mode of data collection is in
factamoderatorofcontexteffects,aviewthatcanprovideananswertothequestion
concerningtheoriginofthedatacacophonyinmixed-modecomparativestudies.
1 Introduction
Research on the comparability of data obtained using different data collection modes in
surveys is often associated with recent developments in quantitative social research. The
falling response rates in traditional self-administered paper, face-to-face and telephone
surveys together with the ever greater time and budget constraints create challenges for
thetraditionalsurveyapproach. Asaresultandgiventhemanypossibilitiesdigitalisation
offers, it is not uncommon for traditional data collection modes in both longitudinal and
cross-sectionalstudiestobecomplementedwithdatacollectedthroughtheInternet. After
all, this is a way of reducing both the cost and duration of the survey fieldwork process
(DillmanandChristian,2005;Jäckle,Roberts,andLynn,2010).
Yet, in reality, these are not new developments. Market research agencies have been
mixingmodesofdatacollectionsincethe 1930safter discovering thattelephonesurveys
provide a faster and more cost-effective way of obtaining data in larger American towns
(Karol, 1937). In 1944, Deming factored “Different kinds and degrees of canvas” as
one of the sources of errors in surveys. He also noted that “cheaper methods” such as
telephone (nowadays the Web) may hold greater utility once the comparability of data
collectionmodesisobtainedthroughmoreextensiveresearch(Deming,1944). Regarding
research on the comparability of data obtained using different modes, already in 1963
Cannell and Fowler developed a set of hypotheses on how the interviewer’s presence
affectsarespondent’smotivationandontheunderreportingofpotentially“embarrassing”
data(CannellandFowler,1963).
Seventy years after Deming’s article, great progress has been made on systematis-
ing measurement errors in different data collection modes (de Leeuw and Berzelak in
Wolfetal,2016;Berzelak,2014;Vannieuwunhuyze,Loosveldt,andMolenberghs,2010;
1
FacultyofSocialSciences,UniversityofLjubljana,Ljubljana,Slovenia;may.dousak@fdv.uni-lj.si

2 M. Doušak
Roberts,2007),althoughthereisstillageneralconsensusamongtheresearchcommunity
that the different modes do not provide equivalent data and therefore combining modes
shouldbeavoidedwherepossible(MartinandLynn,2011).
2
In addition, the advent of connected portable devices such as smartphones and wear-
ables that store vast amounts of users’ information online and give researchers an op-
portunity for continuous “big data” collection raise questions about the need for survey
modeswhichwouldpresentanalternativetocostlyandlengthyface-to-faceorlatelyeven
mailed paper and pencil questionnaires (Mayer-Schönberger and Cukier, 2013). Com-
binedwithtimeandbudgetconstraints,respondents’reluctancetoparticipatealongwith
greatercompetivenessfromprivatemarketingandsurveyagencies,thisforcesresearchers
intostreamliningsurveydesignsandseekingnewapproachestocollectingdata.
This explains why, in practice, the choice of a combination of data collection modes
or new approaches is sometimes all that is available to researchers in order to continue
with their research. To ensure high quality data, researchers must carefully assess all
possible measurement specifics of different data collection modes and sources of bias
(Jäckle,Roberts,andLynn,2010). Forthispurpose,weproposeanewviewontheabove
phenomenon. Based on evidence stemming from both theoretical and empirical sources,
weconceptuallydiscussthethesisthatthemodeofdatacollectionmoderatesthecontext
effects.
2 AboutContextEffects
Before digging deeper into mode effects and their relationship with context effects, we
considerdifferentauthors’viewsoncontexteffectsbysummarisingcertainkeyviewson
contexteffectsfoundintheliterature.
3
Social methodologists typically refer to them as intra-survey stimuli that can have a
noticeable effect on the measurement (Schwarz and Strack, 1991). The most evident is
the context given in the question wording: asking about one’s attitude to abortion in the
context of saving a mother’s life might bring different responses to the same question in
thecontextofunwantedchildren(TourangeauandRasinski,1988).
In addition to this more obvious example, various experiments show that prior ques-
tions or topics can also influence responses to later questions (a phenomenon called the
question-order effect) (Hafner-Fink and Uhan, 2013). Tourangeau and Rasinski’s (1988)
four-stage response process in attitude measurement clearly explains this bias caused by
priortopicsinasurvey. Accordingtotheirdefinition,respondents:
1. “firstinterprettheattitudequestion,determiningwhatattitudethequestionisabout”,
2. “retrieverelevantbeliefsandfeelings”,
3. “applythesebeliefsandfeelingsinrenderingtheappropriatejudgement”
2
While this is true for most surveys, researchers sometimes combine the interviewer-administered data
collection mode (e.g. CAPI) with the self-administered mode for part of the questionnaire (e.g. drop-off
questionnaire, CASI) for sensitive topics where the interviewer’s involvement might have an influence on
therespondent(deLeeuw,2005).
3
A discussion of all of the context effects, their causes and consequences lies beyond the scope of this
article;weonlyrefertosomeofthecontexteffectsneededtoconveyourmessage.

Survey Mode as a Moderator ... 3
4. “usethisjudgementtoselectaresponse”(TourangeauandRasinski,1988: 300).
Ifthe“attitude”andthusalso“relevantbeliefsandfeelings”havealreadybeendeter-
mined and retrieved by prior items in a survey, they could be retrieved in the context of
something which may have a great influence on their judgement process. Tourangeau et
al. (2000) later expanded the four-stage response process beyond attitude measurement
only, although they still discussed the attitudinal questions in a separate section due to
their greater measurement instability. There they explicitly noted “minor question word-
ingororder”(Tourangeauetal.,2000:13)asoneofthereasonsforshiftsinmeasurement.
Whilethereisnoreasonnottoapplysuchideastofactualtypesofquestions,herewelimit
thediscussionsolelytoattitudinalquestions.
The example of a question on one’s attitude to abortion could also have an impact on
more general questions later in the survey as the term “abortion” has already been put
into context. Schuman and Presser did a split-ballot experiment on this very topic and
demonstrated a 13% point difference in answering “yes” to the same question “Do you
thinkitshouldbepossibleforapregnantwomantoobtainalegalabortionifsheismarried
and does not want any more children?” due to the question-order effect (Schuman and
Presser,1996).
Holyk also explained context effect in terms of priming as psychologists refer to it
(Holyk in Lavrakas, 2008). By priming, the words that are semantically or associatively
relatedtopriortermsinaquestionnairearerecognisedfasterthanotherwords. Togivean
example, if “doctor” is mentioned in previous questions, the word “nurse” is recognised
faster. In this case, interpretation and determination of what the question is about occur
more impulsively and may be given less thought and judgement. For some authors, es-
pecially social methodologists, context effects come only from within the questionnaire
itself (intra-survey stimuli). Besides question wording, question order and priming men-
tionedabove,intra-surveystimuliinclude(butarenotlimitedto)responsescaleselection,
serialpositioneffect(primacy,recency),andothers(Lavrakas,2008).
In contrast, Bradburn (1991) pointed out that context effects consist of much more
than just question-order effects or even the questionnaire itself. He argued that “every-
thing that occurs during an interview has the potential to influence the answers given by
respondents to the questions asked by the interviewer” – a view that also accounts for
extra-survey stimuli (Bradburn, 1991). While some readers might argue that his choice
of wording implies the presence of the interviewer, the author also discusses mail ques-
tionnaires. By explicitly mentioning nonsemantic information during an interview such
as “the social context within which the interview takes place”, Bradburn takes a broader,
psychologist’s viewpoint on the survey context. Due to such different views on the con-
textofsurveyinterviews,SchwarzandSudmanevencontendedthatcognitiveresearchers
andsocialmethodologistsdonotspeakthesamelanguage(SchwarzandSudman,1991).
Giventhatthemostcommonlyrecognisedmodelforthecognitiveprocessesunderly-
ingasurveyconversation(Tourangeau,RipsandRasinski,2000)aswellasitsdescendant
model of satisficing (Krosnick, 1991) both come from psychologists, we suggest taking
a wider view of context effects in surveys, similarly to Callegaro, Lozar Manfreda and
Vehovar(2015).
Underthisbroaderview,stimulicanalsocomefromoutsidethesurveyquestionnaire
itself (extra-survey context effects). Examples of such stimuli include (among others)

4 M. Doušak
time constraint and external distractions such as looking out for children or driving a car
while responding to an interview as the respondent is not completely engaged in the sur-
vey response process. According to this view, even the current political situation and the
respondent’sculturalbackgroundmayhaveaninfluenceontheirresponses. Thepresence
ofotherpeoplecanalsoconsiderablyinfluenceresponsestoquestionsonsensitivetopics
–imagineaskingateenagerabouttheirexperiencewithmarijuanainthepresenceoftheir
parents. When talking about the presence of other people, social conformity should be
mentioned(Kelman,1958). Askingaboutone’sattitudetoabortionataChristiangather-
ingwouldseeresultsaspredicted. Eventheinterviewer’spresencecanimpactresponses
suchthattherespondentover-reportssociallyacceptablebehaviourandunder-reportsless
socially acceptable behaviour, what is called “social desirability bias” (Roxas and Lind-
say,2012;ChangandKrosnick,2010;Sakshaug,Yan,andTourangeau,2010;Tourangeau
and Yan, 2007). In order to reduce the cognitive burden of providing accurate and con-
sistent answers, the respondent might also answer in terms of satisficing the interviewer
(Krosnick,1999).
Theinterviewer,specifically,canalsoberegardedasasourceofcontexteffect. Name-
ly, an important aspect of the interviewer’s presence is their conversational implicatures
4
(Grice, 1974). Without an interviewer, the respondent hears (pre-recorded) or only reads
the question literally as it is written. While the interviewer should read the questions
literally as they are written in the questionnaire, he or she does indeed have an influence
on the flow of conversation. Grice (1974) states that dialogues should comply with four
maxims:
• quantity(beasinformativeasrequired–butnotmore)
• quality(donotspreaduntrueorunconfirmedinformation)
• relation(berelevant)
• manner (it is not about the content but about the way the information is given in
termsofambiguities,obscurities,prolixityetc.) (Grice,1974,p. 47).
According to these four maxims, the conversation can indeed change when an inter-
viewer is present compared to self-interviewing (the so-called interviewer effect). Apart
fromsemanticcommunication,thesimilarlyimportantnonsemantic,tacitcommunication
between the respondent and the interviewer should not be disregarded. Both Uhan and
Bradburn used Grice’s framework to argue that interviewer-respondent communication
canbeseenasasourceofcontexteffects(Bradburn,1991;Uhan,1998).
Anothersourceofcontexteffectcanbestimulithatcaninfluencearesponseifsome-
onedoesnothavefirst-handexperienceonthetopic–thisfactisheavilyrelieduponinthe
media, marketing and politics to attract customers or win over sympathisers. One exam-
ple is an individual’s attitude to certain groups of people (based on religion, nationality,
sexualorientation,sportclubloyaltyorothercharacteristics)theyhavenevermetbutare
givenanegativeconnotationinmediareadbytherespondent.
Uhan(1998)goesevenfurtherwithhissystematicdistinctionof“local”and“global”
contexts. The local context encompasses the survey questionnaire and survey situation
4
Indirectly,implicitlyconveyedmeaninginspeech.

Survey Mode as a Moderator ... 5
which not only consist of wording and question-order effects, but also scale selection,
content validity, the respondent’s acquiescence,
5
standardisation of procedures and rel-
evant underlying psychological processes and respondent capability. The latter hints at
both cognitive processes underlying survey interviewing (Tourangeau, Rips, and Rasin-
ski, 2000) as well as the satisficing model (Krosnick, 1999). The idea of local context
encompassingmuchmorethanthequestionnaireitselfisalsosupportedbyotherauthors
whorecognisethecognitiveprocessmodelascapableofexplaining“measurementerrors
duetorespondents”(Callegaro,LozarManfreda,andVehovar,2015,p. 64).
The global context is defined as the specific environment of the interview, formed by
the cultural as well as personality traits of both the interviewer and the respondent who
define the interview’s properties through the constellation of their relationship (Uhan,
1998). The author expands Bingham and More’s (1934) view who state that a survey
interview is a “conversation with a purpose”, and Bradburn’s (1991) view on context
(the “social context within which the interview takes place”) by suggesting that a survey
interviewisaspecificsocialevent.
This definition is particularly valuable while assessing the comparability of survey
data between multiple global contexts among different cultural environments (cross-na-
tional surveys) or different target groups of respondents. Moreover, the global context
hypothesis can also explain differences in data obtained from different data collection
modes such as the interviewer effect as the interviewer brings his/her beliefs, personal
traits,experienceandknowledgeintoalloftheinterviewshe/sheconducts(Uhan,1998).
While one might argue that conversational implicatures should fall within the local con-
text, Uhan explicitly placed them in the global context to emphasise the traits both the
respondentandtheinterviewerbringintothesurveysituation.
Respondent motivation is another aspect of the context within which a survey inter-
view is conducted. A respondent can be motivated by either the content of the survey (a
closeorsalienttopic)oritspresentation(bytheinterviewer,visualappearance,technical
form of the presentation) (Callegaro, Lozar Manfreda, and Vehovar, 2015; Dillman and
Christian,2005). Basedontheabovediscussion,wesystemisesomeoftheselectedcon-
texteffectsbypositioningthemwithintwocategorisationsarisingfromdifferentauthors’
views: theviewofthelocalandglobalcontexteffectsversustheviewofintra-surveyand
extra-surveystimuli,asshowninTable1.
In summary, we see context effects as arising not only from the survey questionnaire
itself, but also from the survey situation, constellation of the interviewer and respon-
dent,aswellastheircultural,politicalandsocialbackgrounds. Thisbroadinterpretation,
whichencompassesviewsheldbybothsocialmethodologistsandpsychologists,provides
the most suitable basis for our main argument, namely that survey mode moderates the
contexteffect. Toarguetherelationshipbetweenthetwo,wewillmorespecificallyfocus
onextra-surveystimulistemmingfrommode-inherentcharacteristics.
5
Acquiescence can be defined as a personality trait (Couch and Keniston, 1960), as well as an effect of
bothsurveycontent(Adorno,etal. 1950;SchumanandPresser,1996)andsurveysituation(socialcontext)
(Campbell,etal.,1960). WhileUhan(1998)explicitlyplaceditinlocalcontextinconnectiontothesurvey
instrument, we refer to it in both the local and global contexts due to the different views on its cause and
nature. Thephenomenonhasbeenstudiedthoroughlyinpsychologicalliteraturewhichtheinterestedreader
mayconsult(DeMaio,1985;Adornoetal.,1950;Campbell,etal.,1960;Uhan,1998).

6 M. Doušak
Table1: Systematisationofglobalandlocalcontexteffectsversusintra-surveyand
extra-surveystimuli
Intra-surveystimuli Extra-surveystimuli
Localcontext • Questionwording
• Questionorder,priming
• Serial position effect (recency,
primacy)
• Acquiescence
• Respondentmotivation
• Scaleselection
• Standardisationofprocedures
• Timeconstraint
• Externaldistractions
• Socialconformity
• Acquiescence
• Respondentmotivation
• Satisficing
• Respondentcapability
• Social context during inter-
view
Globalcontext • Acquiescence • Interviewer’s personality traits
andbeliefs
• Conversational implicatures/
interviewereffect
• Respondent’spersonalitytraits
andbeliefs
• Cultural background (of re-
spondentandinterviewer)
• Politicalandsocialcontext

Survey Mode as a Moderator ... 7
3 Mode Effects and their Relationship with Context Ef-
fects
Themodesemployedincollectingdatainthemeasurementstageareconstantlyevolving,
from paper and pencil, first documented in 1788, through face to face (1912), telephone
(1930s)andmorerecentlymanyothermodestowebandmobile(Callegaro,LozarMan-
freda, and Vehovar, 2015; de Heer, de Leeuw, and van der Zouwen, 1999; de Leeuw,
1992). Notonlyaretheyconstantlyevolving,mostdatacollectionmodesareheretostay
sincenoneoutperformstheothersinallrespects(Couper,2011). Theycanalsobemixed
in either a longitudinal or cross-sectional way, that is, when one mode is superseded by
anotherinalong-runningcomparativesurvey,orwhenmultipledatacollectionmodesare
usedsimultaneously,suchasbytheInternationalSocialSurveyProgramme(ISSP;ISSP,
2010).
Theproblemresearchersfaceisthatthedifferentmodesofdatacollectionsometimes
provide unequal data due to their specific mode-inherent characteristics. This data in-
equityphenomenoniscalledthemodeeffect. Abundantresearchandliteratureisavailable
for the interested reader to consult
6
on the modes of data collection, their main charac-
teristics and the problem of combining them; however, the research on mode effects is
lackinginitsconclusions(deLeeuw,2005).
For the purpose of this article, which seeks to relate mode effects to context effects,
we shall explain the mode effect by narrowing our focus to some of the most commonly
identified mode-inherent characteristics. But before that, let us raise attention to some
effects which are commonly mistaken for mode effects and should be disregarded in our
discussion.
Whenmixingdatacollectionmodes,researcherssometimesgiverespondentsthefree-
dom to choose their own preferred mode for participation. Self-administered paper-and-
pencil questionnaires can be combined with the Web by writing the URL of the Web
questionnaireontopofpage1ofthequestionnaire(LozarManfredaandVehovar,2002).
Asrespondents select themodeoftheirparticipationbasedontheirpreferences, anydif-
ferences between the modes in the resulting data should not be attributed to the mode
effect only but to the selection effect as well. Differences in coverage and nonresponse
of different modes also should not be mistaken for the effect of the data collection mode
perse(Sax,Gilmartin,andBryant,2003;Dillman,etal.,2009;Kaplowitz,Hadlock,and
Levine,2004).
Anotherexampleofeffectwhichcanhowevernotberegardedasamodeeffectrefers
to the following situation. Survey design errors can remain hidden and undetected even
after many years and waves of a survey. Sometimes, they are exposed by a change in
the data collection mode. Dillman and Christian (2005) provide a vivid example of a
survey which identified significantly more “divorced” and “widowed” and fewer “sin-
gle”respondentsafterthedatacollectionmodewasswitchedfromtelephonetotheWeb.
When looking into it, they found a huge shortcoming in the survey design: when the re-
spondentshadbeenaskedabouttheirmaritalstatusoverthetelephone,theyweresimply
6
Examples include (among others) de Leeuw and Berzelak in Wolf et al, 2016; Callegaro, Lozar Man-
freda, and Vehovar, 2015; Berzelak, 2014; Groves, et al., 2009; de Leeuw, Hox, and Dillman, 2008;
Tourangeau,Rips,andRasinski,2000.

8 M. Doušak
askedabouttheirstatuswithoutbeingprovidedthefulllistofoptions. Manywidowedor
divorced respondents answered truthfully that they were single. In contrast, web respon-
dents were given a full list of possible answers so they chose one of options provided.
The inequivalence of the data collected in such cases is not mode-inherent as it could be
avoided in the survey design (disregarding the recency effect we will mention later, the
telephonequestioncouldbereformulatedto“NowIwillreadsomeofthemaritalstatuses
(...) Pleasechoosetheonethatisnearesttoyourcurrentmaritalstatus”).
The above examples are not really mode-inherent effects and can (and therefore sho-
uld) be reduced by the survey design (Hox, de Leeuw, and Zijlmans, 2015; Berzelak,
2014; de Leeuw, Hox, and Dillman, 2008). Thus, some of the most obvious causes of
data inequivalence stemming from different data collection modes can be ruled out if
theyarereducedbythesurveydesign. Whichthenaremode-inherentcharacteristicsthat
createthemodeeffectandcannotbeavoidedevenwithacleversurveydesign?
Whiledifferentauthorsdefinemode-inherentcharacteristicsandhencedifferencesbe-
tweenthemodesintheirownterms,mode-specificcharacteristicsmaybroadlybedefined
asacombinationoftheinterviewer’sinvolvement,thedegreeofcontactwiththerespon-
dent, the channel of communication, the media of communication, the locus of control,
degree of privacy and the presence of computer technology
7
(Berzelak, 2014; Couper,
2011; Groves et al. 2009; de Leeuw, 1992). Authors list such characteristics as mode-
inherent properties that create mode effects and usually cannot be avoided by a clever
survey design or sampling (Berzelak, 2014). While this article cannot fully review all
of them or provide an overall single view on them, we shall briefly examine four of the
characteristics we believe are typically given more thought when designing survey
8
and
relateeachofthemtocontexteffects.
Involvement of the interviewer is certainly the most obvious and generally agreed
mode characteristic (Couper, 2011; Groves, et al., 2009; de Leeuw, 1992; Tourangeau,
Rips,andRasinski,2000).
On the positive side, the interviewer can provide a controlled environment for the
interviewbyarrangingthetimeand(withaface-to-facesurvey)locationoftheinterview
so it can be conducted with the fewest interruptions possible — in contextual terms, he
can ensure the most appropriate social context for the interview (Bradburn, 1991). By
providingthemostappropriatesocialcontext,theinterviewercanconsequentlyinfluence
thelocalcontextwithinwhichtheinterviewiscarriedout.
The interviewer also relieves the respondent of the cognition burden induced by the
questionnaire by taking the locus of control into his domain (for example, by taking care
oftheroutinginthequestionnaire)oreventalkingtolessliteraterespondentswhowould
struggletoreadthequestionnaire. Trainedinterviewerscanalsomotivateandengagethe
respondent (Callegaro, Lozar Manfreda, and Vehovar, 2015). Both respondent capability
andrespondentmotivationarementionedintheprevioussectionandfallwithinthelocal
contextasdefinedbyUhan(1998). Consequently,themodeeffectduetotheinterviewer’s
presencecanmoderatethecontextregardingrespondentscapabilityandmotivationwhen
understoodintermsofBradburn(1991)andUhan(1998).
Despite being a professional, an interviewer can sometimes influence a respondent
7
Foramorein-depthoverviewofmodecharacteristics,seeBerzelak(2014),pp. 32–36.
8
E.g. bydecidingontheinterviewerinvolvement,useofcomputeror/andshowcards,etc.

Survey Mode as a Moderator ... 9
via conversational implicatures (see the section above). This is a global context effect
whichisonlypresentinsurveymodesthatuseaninterviewer. Somerespondentsfeelthe
interviewer is a representative of authority (Uhan, 1998). Both the traditional and con-
temporarypsychologicaltheoryofsocialinfluenceexplaintheprocessesoccurringinthe
interviewer’spresence,suchascomplianceandsocialdesirability. First,Kelman’s(1958)
compliance social influence process
9
can clearly be identified by the interviewer’s pres-
ence. Compliancemeansarespondentprovidesaparticularanswer“toachievefavorable
reaction from another person a group” (Kelman 1958, p. 3). More recently, Friedkin
(1998) built upon the theory of social influence by adding a structural approach to it. In
addition,TourangeauandYan(2007)notethatsocialdesirabilityisquitecommoninsur-
veys (Tourangeau and Yan, 2007). Being influenced by the presence of the interviewer
or a third party on one hand and the sense of privacy in self-administered data collection
modesontheother,socialdesirabilityiswidelyacceptedasamode-inherentcharacteris-
tic (Callegaro, Lozar Manfreda, and Vehovar, 2015; Tourangeau and Yan, 2007; Couper,
2011; Dillman and Christian, 2005). While social desirability is indeed a mode-inherent
effect, Bradburn (1991) also suggests that it is part of the social context in which the
surveyinterviewisconducted.
Basedontheabovearguments,weproposethatinterviewerpresence(involvement)is
amoderatorofthesurveycontextandthusofthecontexteffects.
Thepresenceofcomputertechnologyisidentifiedasamode-inherentcharacteristic
bydifferentauthors(Couper,2011;deLeeuw,2005;Berzelak,2014). Computerpresence
usedtohave(andthereforecanstillhave)animpactonthedataevenwhenusedsolelyby
theinterviewerbecausesomerespondentsfearthatcomputersaremonitoringtheconsis-
tency of their answers (de Leeuw, 1992). While later sources disregard this phenomenon
as today being almost negligible (Snijkers, Hox, and de Leeuw, 1995; de Leeuw, 2002),
the presence of a computer may still change the relations and social setting, thereby af-
fectingthesocialcontext.
Anothercomputer-relatedeffectthatcanonlybeobservedwhenrespondentsactually
have to use it is the level of acquaintance with computer technology (Callegaro, Lozar
Manfreda, and Vehovar, 2015). When a computer is used in telephone surveys, only the
locus of control changes from the interviewer to the computer – the respondent neither
sees nor uses the device but the interviewer’s work is made easier so they can be more
concentratedonmotivatingtherespondent.
Consequently, we find that the computer forms part of the social setting in which
the interview is carried out. When used by the respondent (compared to a paper-based
questionnaire),thecomputermaytakethelocusofcontrolaswellasengageandmotivate
the respondent through visual design, interactivity and other elements (Callegaro, Lozar
Manfreda, and Vehovar, 2015). As the presence of computer technology influences the
socialsettingwhichispartofthe[social]context,italsoinfluencesthecontexteffects.
Another mode-inherent characteristic is the channel of communication (Groves, et
al.,2009;Berzelak,2014). Itcanbeaural(telephone),visual(paperorwebquestionnaire)
oracombinationofboth(face-to-faceinterview). Tourangeau,RipsandRasinskigoeven
further and distinguish channels of presentation and responding (Tourangeau, Rips, and
9
Kelmanidentifiedthreesocialinfluenceprocesses: compliance,identificationandinternalisation(Kel-
man,1958).

10 M. Doušak
Rasinski, 2000). An example of an interview involving different channels of presenta-
tion and responding is IVR where the respondent listens to the question and then enters
the appropriate answer on the telephone keypad. The traditional literature discusses the
dichotomy of oral andliteral communicationwhereby written language is more analytic,
classificatoryandstructuredwhileoralisredundantandaggregative(Ong,1982). Murray
(1988)systematisesthisdichotomyasemotional,personal,context-boundoralcommuni-
cationvs. analyticandsequentialthatissuitableforrationalandcriticalthinkingwritten
text. He (Murray, 1988) further experimented with the computer as the medium of data
collectionandconcludedthatmodeandmediumrepresentchoicesmadeduetothechar-
acteristics of the particular context in which the interaction takes place (Murray, 1988).
Whiledemonstratingthatliteracyandoralityarenotdichotomous,asotherwisesuggested
bythetraditionalliterature, heplacedchannelaswellasmediumofcommunication(de-
scribed here as mode-inherent characteristics) directly in the context of the interaction
(Murray,1988).
In addition, de Leeuw, Hox, and Dillman (2008) suggested that social conventions,
also defined as global context effects by Uhan (1998), can be shaped by the channel of
communication.
Mediumofcommunicationisthelastmode-inherentcharacteristicweusetodemon-
strate that context effects are moderated by the mode of data collection. Primarily influ-
enced by Tourangeau and de Leeuw, Berzelak defined it as an information transmission
medium, which is “the tool (or service) used to transfer questionnaire and answers be-
tween the respondent and the researcher” (Berzelak, 2014, p. 34). He further explains
that it can be “in person, via telephone line, mail or other form of physical delivery, e-
mail, or web” (Berzelak, 2014, p. 34). Returning to his sources, we find the definition
of medium of communication given by de Leeuw (2005) to be one of the most complete
and context-aware. She defines medium-related factors as “social conventions and cus-
toms associated with the media utilized in the survey methods, such as familiarity with a
medium,useofmediumandlocusofcontrol”(deLeeuw,2005,p. 244).
This definition is literally written in terms of context effects by explicitly mentioning
“social conventions and customs”, similarly to Bradburn (1991) and Uhan (1998). In
our opinion, this definition is another formulation of the global contextual view on the
interviewprocess. Asweunderstandit, shemeantthattheuseofanymediumpersecan
causeacertaintracesonthedataassocialconventionsandcustomsvaryamongdifferent
media of communication. Therefore, when the medium of communication is changed,
theglobalcontextwithalltherelatedconventionsandcustomsalsochanges.
Selected mode-inherent characteristics, their (alleged) impact on the survey data as
wellastheirrelationshiptocontexteffectsasdescribedabovearesummarisedinTable2
below.
Let us give an empirical example to illustrate our notion of mode being a moderator
of context. It can be seen from the experience from the European Social Survey (ESS)
10
and the European Values Study (EVS):
11
in order to counteract the differences created
by changing the mode from face to face (F2F) to web, in ESS they had to adapt 85%
10
European Social Survey is a repeat cross-sectional survey on attitudes, beliefs and behavior that is
conductedinover30countrier(ESS,2017).
11
European Values Study is a repeated cross-section survey on basic human values that is conducted in
over45countries(EVS,2017).

Survey Mode as a Moderator ... 11
Table2: Systematisationofselectedmode-inherentcharacteristics,their(alleged)impacton
thesurveydataandrelationshiptocontexteffects
Mode-
inherent
characteristics
Keypossibleimpactsondata Relatedcontexteffects
Presence of
the inter-
viewer(degree
of interviewer
involvement)
• provoke socially desirable an-
swers
• causeinterviewerbias
• relieverespondentofcognitive
burden
• provide the appropriate envi-
ronment
• motivatetherespondent
Local context: Frame the social
contextoftheinterview
Global context: Bring inter-
viewer’spersonalitytraits,beliefs
andexperienceintotheinterview
Channel of
communica-
tion (aural vs.
visual)
• provide access to the illiterate
(aural) provide access to the
deaf(visual)
• ease the survey response pro-
cess for some individuals (de-
pending on their characteris-
tics)
Local context: Trigger different
conversationalimplicatures
Presence of
computer
technology
• limit(orprovide)accesstocer-
taintargetgroups
• instigate different social con-
ventions
• reduce random errors due to
locusofcontrol
• demonstrate social segmenta-
tion (access to the computer,
acquaintance)
Local context: Alter the social
settingduringtheinterview
Global context Bias respondent’s
profile by segmenting regarding
access to computer and acquain-
tancewiththecomputer
Medium of
communication
a
• limit(orprovide)accesstocer-
taintargetgroups
• triggerdifferentsocialconven-
tions, reduce (increase) ran-
domerrorsduetolocusofcon-
trol
• demonstrate social segmenta-
tion(accesstothemedium)
Local context: Provoke medium-
specific social conventions and
customs
Global context Bias respondent’s
profile by segmenting regarding
mediumaccess
a
AsproposedbyBerzelak(2014)basedonTougangeauanddeLeeuw’stexts.

12 M. Doušak
of the questions while in EVS they changed the wording of 156 out of 256 questions
(Callegaro,LozarManfreda,andVehovar,2015)toensurethattheequivalenceofthedata
was retained. We find this change in line with Ong’s (1982) dichotomy of classificatory
and structured written vs. redundant and aggregative oral communication and even more
so with Goody and Watt’s (1963) explanation of semantic ratification, by which “the
meaning of each word is ratified in a succession of concrete situations, accompanied
by vocal inflexions and physical gestures, all of which combine to particularize both its
specificdenotationanditsacceptedconnotativeusages”(GoodyandWatt,1963,p. 306).
BoththeESSandEVSempiricalevidenceaswellasGoodyandWatt(1963)demon-
stratethatanintra-surveycontextadjustment(adaptationofthequestionnaire)canprovide
(approximately) equivalent resulting data when changing the data collection mode from
F2Ftoweb. AsshowninTable2,mode-inherentfeaturescanhaveaninfluenceonextra-
survey context effects. Because the change of mode altered the extra-survey context and
thiseffectcanbereducedbyalteringtheintra-surveycontext(questionnaire),wesuggest
thatmodeitselfisamoderatorofcontextandthereforeofcontexteffects.
4 WhyitMatters
The above discussion demonstrates that the mode of data collection inherently affects
the context effects within which the survey interview takes place. From this viewpoint,
researchers could adjust the survey design to mitigate certain local context effects while
explaining some of the “difficult to predict” (Martin and Lynn, 2011, p. 30) effects in
termsofUhan’sglobalcontext. Forillustrativepurpose,considertwoscenariosofcross-
nationalsurveysbeingconductedinthesamecountries:
1. acompletelystandardised(coherentsampling,coordinatedtranslations,singlemode
etc.) cross-nationalcomparativesurvey(suchastheEuropeanSocialSurvey);and
2. alessstandardisedcross-nationalcomparativesurvey(suchastheInternationalSo-
cialSurveyProgramme).
Whileglobalcontextcannotbedirectlycontrolledfor,itcanbecoherentamongsim-
ilar countries that have similar respondent sampling frames and interviewer bases. How-
ever,itcanalsovaryquiteconsiderablyifthecountriesinvolvedaresufficientlydifferent.
In the first scenario, the completely standardised design provides the same canvas (local
context effect) while the local context effect in the second example may change due to
inequivalent translations, the mode change etc. In the second scenario, the countries in-
volved are quite similar (very similar global context) while the local context is different
duetothelessrigoroussurveydesignstandard.
BothlocalandglobalcontextsaswellasthetwoscenariosarevisualisedinPicture1.
Iftheglobalcontextsvarysubstantially
12
(suchasinscenario1inFigure1),providing
thesamelocalcontextmaybefutilesinceirrespectiveofhowsimilarthelocalcontextis,
the results are considerably inequivalent due to the different global contexts. Scenario 2
demonstratestheinversesituation: verysimilarglobalcontextsprovidethefoundationfor
12
E.g.: ifthecountriesinvolvedhavedifferentcultural,religiousorpoliticalvaluesystems.

Survey Mode as a Moderator ... 13
Figure1: Comparisonofthetwosurveyscenarioswiththesamelocalorsameglobal
contexteffects
higherdataequivalencebuttheopportunityismissedduetodifferentlocalcontextintro-
duced by the less rigorous survey design specifications. The two cases produce different
measuredvalues.
The depiction of the two scenarios emphasizes the true importance of the survey de-
sign for the data obtained. While the amount and impact of both contexts cannot be
assessed beforehand without experimentation, based on the two above scenarios we ar-
gue that, when the global context has a substantially greater effect on the data than the
local context (which includes mode), any effort towards rigid standardization such as in
scenario1maybefutile.
Asthisviewcanonlybevalidwhenmodeofdatacollectionisamoderatorofcontext,
wefindthisrelationshipimportant.
5 Conclusion
Based on both empirical and theoretical evidence, we demonstrated that the effects of
data collection mode may not be separate phenomena but an integral moderating part of
context effects. This view could provide the basis for new and more effective techniques
to be developed to ensure data comparability in mixed-mode comparative surveys and
optimaluseofresourceswheresomestandardisationprocedureswouldproveineffective.
There are many questions in need of answers so further empirical evidence and ex-
perimentationisrequiredinordertoassessthedegreeofmoderationandtherelationship
between different modes within standardised contexts before such techniques can be de-
veloped.
Until then, we should take Martin and Lynn (2011) into account who urged caution

14 M. Doušak
whencomparingdataobtainedusingdifferentmodes.
References
[1] Adorno, T., Frenkel-Brunswik, E., Levinson, D., and Sanford, R. (1950): The Au-
thoritarianPersonality.NewYork: Harper.
[2] Berzelak,J.(2014): ModeEffectsinWebSurveys.Ph.DthesisofUniversityofLjubl-
jana,Ljubljana.
[3] Bingham,W.,andMoore,B.(1934): HowtoInterview.NewYork: HarperandRow.
[4] Bradburn, N. M. (1991): What have we learned? In N. Schwarz, and S. Sudman
(Eds.): Context Effects in Social and Psychological Research, 315–323. New York:
Springer-Verlag,
[5] Callegaro, M., LozarManfreda, K., andVehovar, V.(2015): Web Survey Methodol-
ogy.London: Sage.
[6] Campbell,A.,Converse,P.,Miller,W.,andStokes,D.(1960): TheAmericanVoter.
NewYork: Wiley.
[7] Cannell,C.andFowler,F.(1963): Comparisonofaself-enumerativeprocedureand
apersonalinterview: Avaliditystudy.PublicOpinionQuarterly,27(2),250–264.
[8] Chang, L. and Krosnick, J. (2010): Comparing oral interviewing with self-
administered computerized questionnaires: An experiment. Public Opinion Quar-
terly,74(1),154–167.
[9] Couch, A. and Keniston, K. (1960): Yeasayers and naysayers: Agreeing response
setasapersonalityvariable.TheJournalofAbnormalandSocialPsychology,60(2),
151–174
[10] Couper, M. (2011): The future of modes of data collection. Public Opinion Quar-
terly,75(5),889–908.
[11] de Heer, W., de Leeuw, E. D., and van der Zouwen, J. (1999): Metholological
issues in survey research: A historical review. Bulletin of Sociological Methodol-
ogy/BulletindeMéthodologieSociologique,64(1),25–48.
[12] de Leeuw, E. (1992): Data Quality in Mail, Telephone and Face-to-Face Surveys.
Amsterdam: TT-Publikaties.
[13] de Leeuw, E. (2002): The effect of computer-assisted interviewing on data quality:
A review of the evidence. In J. Blasius, J. Hox, E. D. de Leeuw, and P. Schmidt
(Eds.): Social Science Methodology in the New Millennium. Opladen: Leske + Bu-
drich.
[14] deLeeuw,E.(2005): Tomixornottomixdatacollectionmodesinsurveys.Journal
ofOfficialStatistics,21(2),233–255.

Survey Mode as a Moderator ... 15
[15] de Leeuw, E. and Berzelak, N. (2016): Survey mode or survey modes? In C. Wolf,
D.Joye,T.W.SmithandY.Fu(Eds.): TheSAGEHandbookofSurveyMethodology,
142–156.London: Sage
[16] de Leeuw, E., Hox, J., and Dillman, D. (2008): International Handbook of Survey
Methodology.NewYork: TaylorandFrancisGroup.
[17] DeMaio, T. J. (1985): Social desirability and survey measurement: A review. In C.
Turner, and E. Martin (Eds.): Surveying Subjective Phenomena 2, 257–282. New
York: Sage.
[18] Deming,W.(1944): Onerrorsinsurveys.AmericanSociologicalReview,9(4),359–
369.
[19] Dillman, D., and Christian, L. (2005): Survey mode as a source of instability in
responsesacrosssurveys.FieldMethods,17(1),30–52.
[20] Dillman,D.,Phelps,G.,Tortora,R.,Swift,K.,Kohrell,J.,Berck,J.,andMesser,B.
(2009): Response rate and measurement differences in mixed mode surveys using
mail, telephone, Interactive Voice Response (IVR) and the Internet. Social Science
Research,1–18.
[21] ESS. (2017): About the European Social Survey European Research Infrastruc-
ture – ESS ERIC. Downloaded 04. Sep 2017 from European Social Survey:
http://www.europeansocialsurvey.org/about/
[22] EVS.(2017): AboutEVS.Downloaded04.Sep2017fromEuropeanValuesStudy:
http://www.europeanvaluesstudy.eu/page/about-evs.html
[23] Friedkin, N. E. (1998): A Structural Theory of Social Influence. Cambridge: Cam-
bridgeUniversityPress.
[24] Goody, J., Watt I. (1963): The consequences of literacy. Comparative Studies in
SocietyandHistory.5(3),304–345
[25] Grice,H.(1974): Logicandconversation.InP.Cole,andL.Morgan(Eds.): Syntax
andSemantics3: SpeechArts,44–58.NewYork: Elsevier.
[26] Groves, R., Fowler, F., Couper, M., Lepkowski, J., Singer, E., and Tourangeau, R.
(2009): SurveyMethodology.Hoboken,NJ:JohnWileyandSons.
[27] Hafner-Fink, M., and Uhan, S. (2013): Bipolarity and/or duality of social survey
measurement scales and the question-order effect. Quality & Quantity, 47(2), 839–
852.
[28] Hox,J.,deLeeuw,E.,andZijlmans,E.(2015): Measurementequivalenceinmixed
modesurveys.FrontiersinPsychology,6,1–11.
[29] ISSP. (2010): General information. Downloaded 28. Sep 2016 from International
SocialSurveyProgramme: http://www.issp.org/index.php

16 M. Doušak
[30] Jäckle, A., Roberts, C., and Lynn, P. (2010): Assessing the effect of data collection
modeonmeasurement.InternationalStatisticalReview,78(1),3–20.
[31] Kaplowitz,M.,Hadlock,T.,andLevine,R.(2004): AcomparisonofWebandmail
responserates.PublicOpinionQuarterly,68(1),94–101.
[32] Karol, J. (1937): Measuring radio audiences. The Public Opinion Quarterly, 1(2),
92–96.
[33] Krosnick, J. (1991): Response strategies for coping with the cognitive demands of
attitudemeasuresinsurveys.AppliedCognitivePsychology,5(3),213–236.
[34] Krosnick, J. (1999): Survey research. Annual Review of Psychology, 50(1), 537–
567.
[35] KelmanH.C.(1958): Compliance,identificationandinternalizationthreeprocesses
ofattitudechange.ConflictResolution,2(1),51–60.
[36] Lavrakas,P.(2008): EncyclopediaofSurveyResearchMethods.London: Sage.
[37] Lozar Manfreda, K., and Vehovar, V. (2002): Do mail and Web surveys provide
sameresults? Metodološkizvezki,18,149–169.
[38] Martin, P., and Lynn, P. (2011): The Effects of Mixed Mode Survey Designs on
SimpleandComplexAnalyses.London: CityUniversityofLondon.
[39] Mayer-Schönberger, V., and Cukier, K. (2013): Big data: A Revolution That Will
TransformHowWeLive,WorkandThink.NewYork: HoughtonMifflin.
[40] Murray, D. E. (1988): The context of oral and written language: A framework for
modeandmediumswitching.LanguageinSociety,17(3),351–373.
[41] Ong, W. J. (1977): Interfaces of the World: Studies in the Evolution of Conscious-
nessandCulture.London: CornellUniversityPress.
[42] Roberts,C.(2007): MixingModesofDataCollectioninSurveys: AMethodological
Review.London: ESRCNationalCentreforResearchMethods.
[43] Roxas, B., and Lindsay, V. (2012): Social desirability bias in survey research on
sustainable development in small firms: An exploratory analysis of survey mode
effect.BusinessStrategyandtheEnvironment,21(4),223–235.
[44] Sakshaug, J. W., Yan, T., and Tourangeau, R. (2010): Nonresponse error, measure-
menterror,andmodeofdatacollection: Tradeoffsinamulti-modesurveyofsensi-
tiveandnon-sensitiveitems.PublicOpinionQuarterly,74(5),907–933.
[45] Sax, L., Gilmartin, S., and Bryant, A. (2003): Assessing response rates and nonre-
sponse bias in Web and paper surveys. Research in Higher Education, 44(4), 409–
432.

Survey Mode as a Moderator ... 17
[46] Schuman, H., and Presser, S. (1996): Questions and Answers in Attitude Surveys:
ExperimentsonQuestionForm,WordingandContext.London: Sage.
[47] Schwarz, N., and Strack, F. (1991): Context effects in attitude surveys: Applying
cognitivetheorytosocialresearch.EuropeanReviewofSocialPsychology,31-50.
[48] Schwarz, N., and Sudman, S. (1991): Context Effects in Social and Psychological
Research.NewYork: Springer-Verlag.
[49] Snijkers, G., Hox, J., and de Leeuw, E. (1995): The effect of computer-assisted
interviewing on data quality. A review. International Journal of Market Research,
37(4),325–344.
[50] Tourangeau, R., and Rasinski, K. (1988): Cognitive processes underlying context
effectsinattitudemeasurement.PsychologicalBulletin,103(3),299–314.
[51] Tourangeau, R., and Yan, T. (2007): Sensitive questions in surveys. Psychological
Bulletin,133(5),859–883.
[52] Tourangeau, R., Rips, L., and Rasinski, K. (2000): The Psychology of Survey Re-
sponse.Cambridge: CambridgeUniversityPress.
[53] Uhan,S.(1998): Pravainnepravamnenja.Ljubljana: Fakultetazadružbenevede.
[54] Vannieuwunhuyze, J., Loosveldt, G., and Molenberghs, G. (2010): A method for
evaluating mode effects in mixed-mode surveys. Public Opinion Quarterly, 74(5),
1027–1045.