Documents

EJPA2005

Description
EJPA2005
Categories
Published
of 13
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  T. Buchananet al.:Internet Personality Inventory  European Journalof Psychological Assessment   2005;21(2):116–1 28©2005Hogrefe&HuberPublishers Implementing a Five-FactorPersonality Inventoryfor Use on the Internet Tom Buchanan 1 , John A. Johnson 2 , and Lewis R. Goldberg 3 1 University of Westminster, London, UK,  2 Pennsylvania State University,  Where?? , USA 3 Oregon Research Institute,  Where?? , USA Abstract.  A short five-factor personality inventory developed from the International Personality Item Pool (IPIP) was imple-mented as an online questionnaire and completed by 2,448 participants. Following factor analyses, a revised version was createdwith acceptable reliability and factor univocal scales. As preliminary evidence of construct validity, support was found for 25hypothesized links with self-reports of relevant behaviors and demographic variables. In a replication using a different recruitingstrategy to test for differences due to motivational factors, similar results were obtained. This set of scales appears to provideacceptable measures of the Five-Factor Model for use in internet-mediated research. Keywords:  internet, Five-Factor, personality, assessment Inthepastfewyears,internet-orientedpsychologistshavebegun to generate a body of literature on thepotentialthatthe web may offer for many applied and basic investiga-tions (e.g., Birnbaum, 2000; Gosling, Vazire, Srivastava,&John,2004;Krautetal.,2004;Reips&Bosnjak,2001).Among those for whom the potential is greatest are psy-chologistsinvolvedinthepsychometricassessmentofper-sonality attributes via self-report questionnaires.Itisrelativelyeasytopresenttheitemsofapersonalityinventory on a world wide web page, thus permittingrespondents to complete such surveys in a variety of set-tings convenient to them, and permitting the data to beautomatically processed and scored. This extension of traditionalcomputer-basedassessmentsoffersadditionaladvantages. For researchers, there is the opportunity togather information from very large numbers of individ-uals in a manner that is inexpensive and convenient forboth experimenter and participant (Buchanan, 2000,2001). For those providing assessment services to thecommercial sector, there is the potential to develop newmodes of service delivery (Bartram, 1998; Buckley &Williams, 2002). For those who wish to perform assess-ments in the emergent context of behavioral telehealthservices, internet-mediated assessment may becomequite important (Barak & English, 2002; Buchanan,2002). For all of these purposes, there are additional ad-vantages: (1)Thereissome evidencethatindividuals arelikely to disclose more about themselves in online ques-tionnairesthaninface-to-faceinterviews(Joinson,1999;Joinson & Buchanan, 2001) and, thus, they may respondmore candidly to personality questionnaires; and (2) it isreasonable to suppose that ecological validity may beadvanced when people can complete questionnaires intheir own normal environments (e.g., at home or on anoffice computer) rather than in less familiar laboratorysettings (Reips, 2000).However, there are also challenges to internet-medi-ated assessment that must be identified and addressed.Although laboratory settings might be artificial and un-familiar, they do at least ensure a uniform assessmentcontext. On the internet, standardization of and controlover testing situations is lost (Buchanan & Smith,1999a,b; Reips, 2000). Other challenges include issuesrelating to the motivation of respondents (Buchanan &Smith, 1999b), the potential for dishonest or mischie-vous responses(Reips,2000;Schmidt,1997),thelackof appropriatenorms,anypossibleinteractionsbetweentheconstructs being measured and aspects of the mediumusedformeasurement(Buchanan,2002),andothertech-nical and ethical considerations. The importance of eachof these problems may vary with different applicationsofonlineassessment.However,thereisonequestionthatstrikestotheheartofeverypossibleapplicationofonlinepersonalityassessment:Caninternet-mediatedpersonal-ity measures validly assess the constructs they purporttoaddress? DOI 10.1027/1015-5759.18.1.116  European Journal of Psychological Assessment   2005; 21(2):116–128 © 2005 Hogrefe & Huber Publishers  Evaluations Of Online Personality Scales A number of personality scales have been implementedas online instruments, mainly by amateurs but also byprofessional psychologists (Barak & English, 2002; Ep-stein & Klinkenberg, 2001). To date, however, relativelyfew evaluations of the psychometric properties of onlinepersonality tests have been published. Of those accountsthat have appeared in the literature, there are several thatpermit evaluation of the online tests either by inclusionof a paper-and-pencil comparison condition, or by com-parison with findings obtained from some other equiva-lent method. The majority of these reports focus on sin-gle personality constructs rather than broad inventories.In general, the findings from such studies have shownthat online measures can be reliable and valid measuresof the intended constructs (see Buchanan, 2001, 2002,for reviews).However, the picture becomes a little more complexwhen multi-factorial inventories are considered. Robins,Tracy, and Trzesniewski (2001), in an examination of links between self-esteem and other personality traits,used data obtained from an online version of a Big Fivepersonality inventory (John & Srivastava, 1999) in avery large sample. They reported internal-consistencyreliability estimates (coefficient  α ) for each of the fivedimensions that were as high as those obtained with thepaper-and-pencil version of the inventory. Woolhouseand Myers (1999) evaluated a new measure (based on aJungianpersonalitytypology)usingbothpaper-and-pen-cil and internet modes of administration, and similarlyfound that the reliabilities were comparable. However,they also found some differences in the latent structuresfor the two versions of the instrument; the factors onwhich some items had their highest loadings variedacross the two methods. Fouladi, McCarthy, and Moller(2002) compared findings from three questionnairesmeasuring, respectively, attachment to father and tomother; perceived ability to reduce negative mood; andawareness of mood and mood regulation strategies.These were administered either online or in paper-and-pencil format, with traditionally recruited participantsrandomly assigned to conditions. Although psychomet-ric properties and latent structures were in general verysimilar, there were some small differences between theinternet-mediated and traditional versions of the instru-ments (e.g., mean scores on two of the mood-relatedscales differed significantly, there were differences inmeans and variances on a number of items, and therewere some distributional differences). Fouladi et al. ar-gued that although their findings demonstrate that inter-net administration of psychological measures is clearlyviable, they also indicate the need for further evaluationand refinement of such measures.Johnson (2000) created a web-mediated version of anInternational Personality Item Pool (IPIP: Goldberg,1999) representation of the constructs in Costa and Mc-Crae’s Five Factor Model as embodied in their NEO-PI-R(Costa&McCrae,1992).FactoranalysesofJohnson’sdata showed that, broadly speaking, the latent structurewas as expected but that there were some anomalies,with a small minority of the facet-level constructs load-ing most highly on the “wrong” domain construct.The findings from studies that permit an evaluation of the psychometric properties of online personality tests(both single and multiple construct measures) led Bu-chanan (2002) to conclude that internet-mediated testscould be reliable and valid, but that online and offlineversions of the same test “can be equivalent but are notalways identical. One cannot take the psychometricproperties of online instruments for granted, even if theyare direct translations of traditional instruments”(p.150). For confidence that one is using a satisfactorymeasure – especially in these early days of online testing(Robinsetal.,2001)–thepsychometricpropertiesofanyonline test need to be established before any real weightcan be attached to data derived from it.The purpose of the current exercise was to develop anonlineinstrumentwithdemonstrablyacceptablepsycho-metric properties, for use in internet-mediated psycho-logical research. Giventhe current popularity of the FiveFactor Model (Costa & McCrae, 1992) as a representa-tion of some central constructs of personality, it is likelythat an instrument measuring these five dimensions –Extraversion (E), Agreeableness (A),Conscientiousness(C), Neuroticism (N), and Openness to Experience (O) –could be extremely useful for online research. Instrument Selected The instrument chosen was the 50-item IPIP representa-tion of the domain constructs of the Five Factor Model,asexpressed inCostaandMcCrae’s(1992)revisedNEOpersonality inventory (NEO-PI-R). There are a numberof reasons why this particular instrument was selected.First, the NEO-PI-R is a very widely used inventory.There isanextensiveliteratureon theextenttowhichtheconstructs embodied in this inventory relate to variousbehavioral criteria and other phenomena of psychologi-cal interest, and the scales have proven to be useful toolsin a number of applied fields. The scales in the IPIP im-plementation have been shown to correlate highly withthe corresponding NEO-PI-R domainscores,with corre-lations that range from .85 to .92 when corrected for un-reliability (International Personality Item Pool, 2001).The IPIP scales also outperformed the NEO-PI-R ver-sions of the same constructs as predictors of a number of T. Buchanan et al.: Internet Personality Inventory 117 © 2005 Hogrefe & Huber Publishers  European Journal of Psychological Assessment   2005; 21(2):116–128  clusters of self-reported behavioral acts (Goldberg, inpress),althoughthesefindingscomefrom thesamesam-ple as was used to construct the IPIP scales.Second, the IPIP representation (like all IPIP mea-sures)isfreelyavailableinthepublicdomain(Goldberg,1999). With proprietary instruments, there are potentialcopyrightandtestsecurityissuesthatmightpreventtheiruse on the internet. With public-domain instruments,these problems do not arise.Finally, the instrument is relatively short. Web experi-ments are subject to relatively high dropout rates (Musch& Reips, 2000; Reips, 2000), partly due the fact that it iseasier to leave a web experiment than one conducted in atraditional environment. Although there are a number of factorslikelytoaffectdropout,suchasfinancialincentives(Frick,Bachtiger&Reips,2001;Musch&Reips,2000)itis likely that longer questionnaires (such as those that in-cludealongpersonalityinventoryfollowedbysomeotherquestions addressing topics of interest) will lead to largernumbers of people abandoning the study (Knapp & Hei-dingsfelder, 2001). Such attrition is potentially a seriousissue because of the possibility of selective dropout. It islikelythatthoseparticipantswhodropoutearlywilldifferfrom those who continue to the end of the survey in traitssuchasconscientiousnessandpatience.Thismaylimitthegeneralizabilityofthefindingsandbiastheresultsofstud-ies where independent or outcome variables are related tothosecharacteristicsthatmightaffectaperson’slikelihoodto terminate participation early. As a consequence, otherthingsbeingequal,shortscalesaredesirableforuseonline.  Validation Strategy A measure of personality is of little use if it does notpredict any behavioral outcomes. Accordingly, much ef-fort has been invested in discovering the correlates of major personality dimensions, and a number of behav-iors associated with each of the domains in the Five Fac-tor Model dimensions have been documented in the lit-erature. The primary validation strategy adopted in thecurrent study was to ask participants for self-reports of the frequency with which they had engaged in each of those behaviors that have been linked toscores on one ormore of the Five Factor dimensions. Correlations be-tweentheseselfreportsandtherelevantdimensionsmaybe regarded as support for the contention that the scalesmeasure the intended constructs. Self-Reports of Criterion Behaviors Booth-KewleyandVickers(1994)foundthattrafficrisk-taking was negatively related to Agreeableness (A) andConscientiousness (C), and positively to Neuroticism(N). There is also evidence that N is related positively totraffic accidents (Schenke & Rausche, 1979) and C isnegatively related to involvement in driving accidents(Arthur&Graziano,1996).Twobehavioralitemsrelatedtothesefindingswereconstructed:“Hadaspeedingtick-et or fine?” (Behavior 1) and “Been involved in a trafficaccident which was at leastpartly yourfault?” (Behavior2). It was hypothesized that, among those respondentswho have access to cars, N will be positively related tobothofthesebehaviors,whileAandCwillbenegativelyrelated to them.Heaven (1996) found that Agreeableness was nega-tively associated with interpersonal violence for bothmen and women. Therefore, a negative correlation waspredicted between A and the behavioral item “Been in-volved in a fight?” (Behavior 3). Heaven (1996) alsoreported that Conscientiousness was negatively associ-ated with vandalism for both men and women. We maythus expect a negative correlation between C and thebehavioral item “Vandalized something or damaged theproperty of another?” (Behavior 4).In introductory personalitytextbooks,aliking forpar-ties and social situations is often presented as one of thedefining characteristics of the extravert. Goldberg (inpress) found a positive correlation ( r   = .31) between Ex-traversion (E) and the item “planned a party” in a largecommunity sample. We, therefore, expected that Ewould correlate positively with the behavioral item“Planned a party?” (Behavior 5). Eysenck (1976) report-ed that extraverts tended to have more sexual partnersand were more likely than introverts to report more thanone sexual partner in the past year. It was, thus, hypoth-esized that in the current sample a positive correlationwould be found between E and the behavioral item “Hadsexual intercourse with a person other than your current(or most recent) partner?” (Behavior 6).A number of studies have explored the associationsbetween personality traits and smoking.Helgason, Fred-rikson, Dyba, and Steineck (1995) found that extravertssmoked more and found it harder to quit. We, therefore,predictedapositivecorrelationbetweenEandthebehav-ioral item “Smoked tobacco (cigarettes, cigars, pipe)?”(Behavior 7).McCraeandCosta(1997)considerartiststobe“primeexamples of individual high in Openness to Experience”(p.825). Although the number of professional artists islimited,many people – as McCrae and Costa note–haveartistic dispositions or interests, and may engage in pro-duction of artworks at an amateur level. Goldberg (inpress) found that the item “produced a work of art” cor-related positively ( r   = .38) with Openness (O) in a com-munitysample.Accordingly,wepredictedapositivecor-relation between O and the behavioral item “Created a118 T. Buchanan et al.: Internet Personality Inventory  European Journal of Psychological Assessment   2005; 21(2):116–128 © 2005 Hogrefe & Huber Publishers  work of art (e.g.,paintingorsculpture)?”(Behavior 8)inthe internet sample.McCraeandCosta(1997)reportedpositivecorrelationsbetweenOandtheneedsforSentienceandUnderstandingassessed in thePersonality Research Form, and suggestedthat“theintellectualinterestsofopenmenandwomenmaylead them to seek higher levels of education” (p.831). Itwas thus predicted that O would correlate positively withthe behavioral item “Taken a course or class you did nothave to, purely out of interest?” (Behavior 9).In addition, Goldberg (in press) has found relation-ships between scores on the paper-and-pencil version of the scale to be used in this study and various behavioralacts in a large community sample. Some of the highest,which are clearly conceptually related to the dimensionswith which they correlate, were selected for use in thisstudy. We predicted that N would correlate positivelywith the behavioral item “Taken medication for depres-sion?” (Behavior 10), that E would correlate positivelywith the behavioral item “Started a conversation withstrangers?” (Behavior 11), that O would correlate posi-tively with “Attended an art exhibition?” (Behavior 12),that A would correlate negatively with “Made fun of someone?” (Behavior 13), and that C would correlatenegatively with “Let work pile up until just before adeadline?” (Behavior 14). The correlations from Gold-berg (in press) between these dimensions and behaviorswere, respectively, .38, .34, .48, –.32, and –.29. Other Predictions In addition to these behavioral acts, a number of demo-graphic and other variables accessible through self-re-ports are known to be linked to one or more of the fivefactors. If the same associations are found in the currentsample, this may also be interpreted as evidence for con-struct validity.Anumber of sex differences in personality traits havebeen reported (Budaev, 1999). Normative data on thedomain constructs of the NEO-PI-R by Costa and Mc-Crae (1992) show some such differences. The mostmarkedarewithNandA,withvaluesofGlass’s d  aroundthe magnitude corresponding to a medium effect size inCohen’s (1992) terms; men score lower on both N ( d   =.38) and A( d   = .55) than do women 1 . One would expectto find the same pattern here.As already noted, McCrae and Costa (1997) suggestedthat people high on O will be more likely to seek out edu-cational experiences. In addition to the education-seekingbehavior(Behavior9),onemightexpecthighOscorerstohavehighereducationallevels.Similarly,thereisevidence(De Raad & Schouwenburg, 1996) that people high in Care likely to have higher levels of education. Therefore,onemighthypothesizethatamongparticipantswhodonotreport their occupations as “student” (thereby implyingthat theirformal education has finishedatleasttemporari-ly), both O and C will be positively associated with thehighest level of formal education completed.Negative associations have been reported between Nand job satisfaction, with higher scorers tending to beless happy with their jobs, whatever those jobs may be(Perone, DeWaard, & Baron, 1979). Although job satis-faction may be bestcharacterizedasmultidimensionalinnature (e.g., satisfaction with elements such as salary,relationships with workmates,orworking environment),it seems reasonable to predict that people scoring highon N will report less satisfaction with their jobs in gen-eral. Associations between N and health were also ex-pected. Costa and McCrae (1992) suggested that “N is apotentpredictorofsomaticcomplaints”(p.35),andVoll-rath, Knoch, and Cassano (1999) reported that peoplehigh on N tend to worry more about their health. Finally,links between N and mental health are also well docu-mented (e.g., Miller, 1991). It was, therefore, predictedthat N would be negatively correlated with self-reportsof both physical and mental health.In summary, then, the purpose of Study 1 was to es-tablish whether an internet-mediated version of an IPIPFive Factor Inventory was psychometrically acceptablein terms of factor structure and reliability. In addition, apreliminary assessment of its validity was conducted bytesting 25 hypothesized relations between the five fac-tors andself-reportsofbehaviors,demographic,andoth-er variables. Study 1: Materials and Methods Materials An internet version of the short IPIPinstrument measur-ing the domain constructs of the Five Factor model wascreated as a series of web pages. All pages were dynam-ically generated in the scripting language Perl. Thescripts generated the code to display the pages in theparticipant’sbrowser,andtologtheirresponses(andoth-er contextual information: date, time, browser type, andinternet address of the computer they were using) to ourdata-files.Participants first saw a page with a brief descriptionof the inventory, explaining that it was part of a researchT. Buchanan et al.: Internet Personality Inventory 119 © 2005 Hogrefe & Huber Publishers  European Journal of Psychological Assessment   2005; 21(2):116–128 1 These calculations are based on the data presented by Costa and McCrae (1992) in Appendix B, Table B–1 “Means and Standard Deviationsfor Form S for Adults” (p.75).
Search
Similar documents
Tags
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks