psy444

psy615week4dis1articledownload.pdf

Correcting Big Five Personality Measurements for Acquiescence: An 18-Country Cross-Cultural Study

BEATRICE RAMMSTEDT*, CHRISTOPH J. KEMPER and INGWER BORG GESIS – Leibniz Institute for the Social Sciences, Mannheim, Germany

Abstract: For groups of persons with low or medium levels of education, Big Five personality scales typically yield scores that poorly replicate the idealized Big Five factor pattern. On the basis of representative samples of German adults, Rammstedt et al. have demonstrated that correcting each person’s score for acquiescence eliminates this problem. In the present 18-country study using large samples representative of each country’s adult population, we found that, in all cases, correcting for acquiescence did indeed improve the congruence of factor loadings with an idealized Big Five pattern. However, although this correction led to acceptably high correspondence levels in all countries classified as individualistic, this was not always true for non-individualistic countries. Possible reasons for this finding are discussed. Copyright © 2012 John Wiley & Sons, Ltd.

Key words: Big Five; cross-cultural; acquiescence; factor structure; education

Acquiescent responding has a distorting impact on the psychometric quality of questionnaire data (e.g. Rammstedt, Goldberg, & Borg, 2010; Rammstedt & Kemper, 2011; Soto, John, Gosling, & Potter, 2008). Acquiescence refers to the tendency of an individual to consistently agree to questionnaire items, regardless of the content of the items (Jackson & Messick, 1958; Javeline, 1999). Results of previous studies indicate that such ‘yea-saying’ appears to be more frequent among persons with lower educational levels (e.g. Ayidiya & McClendon, 1990; Narayan & Krosnick, 1996; Rammstedt & Kemper, 2011; Rammstedt et al., 2010). There are several hypotheses aiming to explain this educational difference. The most plausible explanation is that the less educated are not as used to thinking abstractly and to considering the hypothetical but are rather tied more to the concrete and immediate (Flavell, Miller, & Miller, 1993; Toomela, 2000, 2003a, 2003b). These competencies, furthermore, are assumed to be related to a person’s likelihood to arrive at meaningful judgements describing himself in relatively abstract psychological terms. Hence, judgements by less educated individuals should be more affected by systematic response biases such as acquiescence (cf. Goldberg, 1963; Soto et al., 2008).

Factor solutions may be blurred by acquiescence, because individual differences in acquiescence will inflate correlations between items that measure unrelated constructs but are keyed in the same direction and will deflate correlations between items that measure the same construct but are keyed in the opposite directions. Thus, individual differ- ences in acquiescence significantly decrease the factorial validity of questionnaires. The most established factorial

model in personality consists of the so-called Big Five factors of personality (cf. De Raad, 2000; Goldberg, 1990; John, Naumann, & Soto, 2008) aiming to describe a person’s personality structure on the most global level. These global dimensions are interpreted and commonly labelled as Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness to Experience.

Recent studies (Rammstedt & Kemper, 2011; Rammstedt et al., 2010) suggest that the structure of item sets assessing the Big Five personality factors is sensitive to effects of acquiescent responding. In samples of persons with low and medium education and thus in samples more prone to acquiescence, the Big Five structure emerged only in a blurred way, whereas in samples consisting of respondents with a high level of education and thus in samples less prone to acquiescence, the Big Five replicated with textbook-like clarity. This educational effect could be substantially reduced by correcting for acquiescent response bias. When this bias was controlled for, the Big Five factor structure became much clearer, reaching good levels of congruence with idealized Big Five loading patterns both for high-educated respondents and for low-educated and medium-educated respondents.

As the effect of acquiescence blurring the Big Five factor structures has only recently been identified, the empirical evidence for its generalizability is still scarce. The two studies that have been conducted so far are both based on representative samples of German adults. Both show (i) that acquiescent responding is significantly and substantially more pronounced in low-educated and medium-educated groups than in high-educated groups; (ii) that the Big Five factor structure is blurred by acquiescence; and (iii) that statistically controlling for acquiescence clearly improves the congruence of factor structures with ideal patterns, thereby yielding solutions that have at least an acceptable good fit to the ideal Big Five structure. As

*Correspondence to: Beatrice Rammstedt, GESIS – Leibniz Institute for the Social Sciences, PO Box 12 21 55, D-68072 Mannheim, Germany. E-mail: [email protected]

European Journal of Personality, Eur. J. Pers. 27: 71–81 (2013) Published online 11 December 2012 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/per.1894

both these studies are based on German samples, one can ask to what extent this effect can also be replicated in other cultures — in cultures more or less comparable with and in cultures that differ greatly from the German culture.

The findings described earlier are supported by other studies. Several earlier studies (Krosnick, Narayan, & Smith, 1996; McClendon, 1991; Meisenberg & Williams, 2008; Mirowsky & Ross, 1991; Winkler, Kanouse, & Ware, 1982) have suggested that acquiescent responding is related to education. Thus, as a default hypothesis, we expect the following to hold true in general:

H1: In all cultures, the average level and the variability in acquiescent responding are greater in less educated respondents than in more-educated respondents.

Our own previous studies based on German samples suggest that interindividual differences in acquiescent responding blur the Big Five factor structure in educationally heterogeneous samples. In particular, samples with lower- educated respondents seem to be affected. By subtracting each individual’s mean item score from each single item score in a balanced battery of 10 Big Five items, we statistically controlled for interindividual differences in acquiescence in these samples, leading to a much clearer Big Five factor structure. The assumption that differences in acquiescence blur the Big Five factor structure is also supported by Soto et al. (2008) who showed that children and adolescents exhibit greater interindividual variability in acquiescence than adults and that this negatively impacts the factorial structure of their personality ratings. Control- ling for acquiescence (by ipsatizing the data) improved the recovery of the intended Big Five structure. Hence, we hypothesize that

H2: Statistically controlling for acquiescence generally leads to a factor structure that is more congruent with an idealized Big Five structure.

Our own studies (Rammstedt et al., 2010; Rammstedt & Kemper, 2011) suggest that the negative impact of acquies- cence variability on the factor structure is more pronounced among low-educated and medium-educated respondents than among high-educated respondents. Therefore, we hypothesize that correcting for acquiescence leads to corre- spondingly greater improvements in factor congruence in the case of lower-educated and medium-educated respon- dents. As there is no apparent reason why this should hold for German samples only, we hypothesize for all countries that

H3: Samples of low-educated and medium-educated persons exhibit Big Five factor structures that are less congruent with ideal Big Five patterns than do samples of high-educated persons.

H4: Statistically controlling for acquiescence variability leads to greater improvements in the congruence of empirical and ideal factor structure for low-educated and medium-educated persons than for high-educated persons.

Recently, research on response styles in different cultures has received considerable attention (e.g. Clarke, 2001; Diamantopoulos, Reynolds, & Simintiras, 2006). However, almost all of these studies have focused on cross-cultural differences rather than on cross-cultural generalizability. There is some evidence that cultures differ in their propensity for acquiescent responding as reflected in mean-level differences (e.g. Javeline, 1999; Johnson, Kulesa, Cho, & Shavitt, 2005). This appears to be especially true for the individualism–collectivism dimension — a dimension that is sometimes also regarded as the ‘deep structure’ of cultural differences (cf. Greenfield, 1999). As a possible reason for this effect, Javeline (1999) argues that ‘yea sayers’ are more likely in ‘polite’ societies.

With reference to the quality of the emic Big Five structures, previous studies also indicate differences between individualistic and collectivistic — or non-individualistic — cultures. Not surprisingly, as the Big Five structure was originally identified in an individualistic country — the USA — replications of the Big Five in individualistic countries show greater congruence with this original solution than replications from non-individualistic cultures (cf. Church & Katigbak, 2000; Guanzon-Lapena, Church, Carlota, & Katigbak, 1998).

As educational bias in acquiescence and the effectiveness of correcting scores for acquiescence have been found in Germany, a country typically regarded as individualistic, we hypothesize as an alternative to the universal hypotheses proposed earlier that

H5: The educational bias in acquiescence, and the effectiveness of correcting scores for acquiescence, will be replicable primarily in countries with a culture similar to Germany’s, that is, in individualistic cultures.

H6: The congruence of the Big Five factor structures will be greater in individualistic cultures than in non-individualistic cultures.

The present study aims to test these six hypotheses. For that purpose, data representative of the adult population of 18 countries that differ in terms of their political and economic situation and their cultural values were investigated.

METHOD

Samples and procedure

Our analyses are based on the data of Round 2005 of the International Social Survey Programme (ISSP) (www.issp. org; http://dx.doi.org/doi:10.4232/1.4350; Haller, Jowell, & Smith, 2009). A short measure of the Big Five1 was assessed

1Because of the severe time limitations in large-scale surveys such as the ISSP, in which each individual’s response is counted in monetary equiva- lents, these surveys are unable to assess full-scale personality scales contain- ing 50 to 250 items.

72 B. Rammstedt et al.

DOI: 10.1002/per

http://dx.doi.org/doi:10.4232/1.4350

in 18 of the 31 countries participating in this round. Data of these countries are analysed in the present study.

Without exception, the ISSP surveys were all based on representative samples of the countries’ adult populations (age ≥ 18 years).2 The sampling procedure varied across countries. In several countries, register-based samples were drawn (where possible); in other countries, a household sampling procedure was used. In the latter case, households were selected randomly on the basis of complete block lists. Similarly, respondents within the households were randomly selected from a full list of household members. Full details of the sampling procedures in the different countries are presented in the methods report of the ISSP 2005 (Scholz, Harkness, & Faass, 2008).

Sample size per country varied from N = 921 in Japan to N = 2171 in Taiwan. Overall, the realized sample consisted of N = 25 509 respondents. Details of the sample composition per country are given in Table 1.

ISSP questionnaires are administered as face-to-face interviews or in a self-completion format. Participation was voluntary and not usually financially rewarded — although some countries offered small incentives in order to enhance participation.

Measures

Education Levels of educational attainment were assessed in all countries on the basis of the national education system. Afterwards, the ISSP researchers harmonized these national data by mapping them into six categories: ‘no formal qualifi- cation’ (0), ‘lowest formal qualification’ (1), ‘above lowest qualification’ (2), ‘higher secondary completed’ (3), ‘above higher secondary level’ (4), and ‘university degree

completed’ (5). These categories were defined with reference to the International Standard Classification of Education (ISCED) 1997 levels. In order to have a sufficient number of respondents in each group, we collapsed these six categories into two educational groups on the basis of the ISCED levels3: low or medium education (≤ISCED level 3), which comprises respondents with a level of education lower than that of university students typically investigated in personality research; and high education (≥ISCED level 4). Table 1 shows the distributions of the samples in the low-educated and medium-educated group and in the high-educated group. Depending on the education system, these distributions differ markedly across countries. They range from 40% low-educated and medium-educated and 60% high-educated respondents in Russia to 90% low- educated and medium-educated and 10% high-educated respondents in the Czech Republic. These differences might be due to differences in non-response bias among the countries with some countries showing greater effects for education than others. These differences might also be a consequence of the somewhat arbitrary classification into low or medium and high education, strongly reducing variance and differences among the different educational systems.

Cultural values Country-level scores on an individualism–collectivism scale were taken from Hofstede (2001; see also Hofstede & Hofstede, 2004) for the 18 ISSP countries used in this study. Countries were classified into three groups — high (=3), medium (=2), and low (=1) individualistic — based on their individualism–collectivism scores. For two countries (the Dominican Republic and Latvia), no country- level scores were available. On the basis of the ratings of comparable countries, we assigned imputed individualism– collectivism scores to these two countries. As all other Caribbean countries are classified by Hofstede as having a low level of individualism, we assumed this to hold true for the Dominican Republic as well. We classified Latvia as medium individualistic because Estonia, the neighbouring Baltic country, is rated this way. On the basis of the classification by Hofstede, eight of the 18 ISSP countries were classified as high individualistic cultures, five as medium individualistic cultures, and five as low individualistic cultures (cf. Table 2).

The 10-item Big Five Inventory The 10-item Big Five Inventory (BFI-10; Rammstedt & John, 2007) is an abbreviated version of the well-established Big Five Inventory (BFI; John, Donahue, & Kentle, 1991; see also Benet-Martínez & John, 1998; John et al., 2008; for the German version, see Lang, Lüdtke, & Asendorpf, 2001; Rammstedt, 1997), consisting of 10 of the 44 standard BFI items. It assesses the Big Five with two items per factor, one keyed in the positive and one in the negative direction.

2France had a lower age cut-off of 15 years; Japan had a lower cut-off of 16 years; and Flanders reported an upper age cut-off of 85 years.

Table 1. Sample composition for the 18 countries separately

Country N %

females Mean age

% low/medium education

USA 1518 53.4 47.1 48.9 Germany 1701 51.4 49.4 85.4 Ireland 1001 57.8 46.7 65.9 New Zealand 1309 54.0 47.9 51.9 France 1620 54.8 45.3 54.1 Denmark 1598 52.6 46.9 50.3 Switzerland 1078 53.0 49.9 73.3 Flanders 1338 50.5 46.2 70.5 Latvia 1067 59.9 45.3 76.5 Russia 1605 54.5 44.6 39.9 Czech Republic 1226 59.5 45.6 89.6 Israel 1184 51.9 46.3 56.7 Japan 921 53.8 52.7 67.4 Philippines 1200 50.0 41.8 65.7 Mexico 1401 51.6 37.2 83.3 Taiwan 2171 49.1 44.2 63.9 South Korea 1613 54.4 44.6 55.9 Dominican Republic

1958 52.2 37.9 81.9

3Previous studies by Rammstedt et al. (2010, 2011) based their analyses on three educational groups. Results in both studies indicated that the low- educated and medium-educated groups behaved quite similarly and could be clearly differentiated from the high-educated group.

Correcting Big Five measurements for acquiescence 73

DOI: 10.1002/per

In accordance with the response format used throughout the ISSP questionnaire, all items used 5-point Likert-type response options ranging from fully agree to fully disagree. Although this response scale differs in its orientation from the standard BFI-10 format (fully disagree to fully agree), a previous study showed that reversing the direction of the response scale does not change the quality of the ensuing responses (Rammstedt & Krebs, 2007). All 10 items were re-coded for the present analyses so that fully disagree was scored as 1 and fully agree as 5.

Analyses

Acquiescence An acquiescence index was computed for each respondent on the basis of the 10 BFI items. The BFI-10 has two balanced keyed items for each of the Big Five factors. For example, Extraversion is assessed by the items ‘I see myself as someone who is reserved’ and ‘I see myself as someone who is outgoing, sociable’. If an individual responds consistently to the content of the two items, the responses to these items should be symmetrical about the answer scale’s middle category (coded as ‘3’). High acquiescence, by contrast, should result in a mean rating greater than 3 based on a response scale coded from 1 to 5, indicating that agree- ment to the positive item is stronger than rejection of the negative item. Low acquiescence, namely the tendency to generally disagree with items (i.e. ‘nay-saying’), should result in a mean rating lower than 3. Because there are five

item pairs, we used the mean overall items as a measure of each individual’s acquiescent response tendency. In order to test whether acquiescence affected all five dimensions to the same degree, we computed acquiescence scores for each dimension (as a mean across both items per dimension) and investigated the interrelations of these scores. These correlations range from .08 to .21 with an overall mean of .15.

Controlling for acquiescence One common technique to remove variance due to response tendencies from personality questionnaires is ipsatization (Clemans, 1966; Cunningham, Cunningham, & Green, 1977; Fischer, 2004; Ostendorf, 1990; Ten Berge, 1999). Depending on the theoretical and conceptual interest of the researcher, standardization of item scores may involve adjustments using each person’s means, standard deviations, or both. If researchers expect acquiescent response bias and if this bias is their primary interest, the use of means to adjust item scores is indicated (cf. Fischer, 2004). Ten Berge (1999) also proposed as an alternative approach to use linear regres- sion to partial variance due to acquiescence. This approach differs to the subtraction of the mean in that it only affects the variables to the extent that they correlate with the mean (Ten Berge, 1999, p. 97). However, as our aim is to apply a method of controlling for acquiescence that is not sample dependent, we chose to use the established subtraction method instead. Therefore, we subtracted each respondent’s mean response across all BFI-10 items from his or her score on each single item. To control for

Table 2. Mean and variability of the acquiescence scores and percentage of acquiescent respondents for the 18 countries (total sample and each educational group)

Country Individualism

Acquiescence score

% of low acquiescence

(<2.8)

% of high acquiescence

(>3.2)

Total Low/medium education High education

M SD M SD M SD

USA 3 3.32 .31 3.34 .32 3.30† .30} 3.0 57.5 Germany 3 3.22 .35 3.23 .35 3.16† .35 7.5 44.3 Ireland 3 3.31 .31 3.30 .31 3.31 .32 2.0 53.1 New Zealand 3 3.29 .37 3.33 .40 3.26† .33} 3.9 50.6 France 3 3.24 .45 3.29 .53 3.15† .34} 7.5 39.3 Denmark 3 3.20 .39 3.24 .42 3.15† .34} 9.3 39.9 Switzerland 3 3.22 .28 3.23 .29 3.19† .26 3.3 43.5 Belgium (Flanders) 3 3.15 .32 3.15 .34 3.15 .26} 7.7 33.6 Latvia 2 3.17 .36 3.14 .36 3.26{ .35 11.4 41.0 Russia 2 3.46 .43 3.46 .45 3.46 .43 3.9 69.7 Czech Republic 2 3.16 .38 3.15 .38 3.24{ .37 13.4 39.8 Israel 2 3.52 .38 3.55 .41 3.48† .35} 2.1 76.9 Japan 2 3.00 .46 3.00 .46 3.00 .47 26.1 27.0 Philippines 1 3.25 .47 3.22 .49 3.30{ .41} 13.3 50.0 Mexico 1 3.55 .49 3.56 .50 3.54 .47 3.4 67.2 Taiwan 1 3.15 .33 3.11 .33 3.21{ .31 9.6 35.8 South Korea 1 3.15 .36 3.14 .39 3.15 .31} 11.4 35.6 Dominican Republic 1 3.45 .36 3.44 .36 3.49{ .36 2.8 70.1 Total mean 3.27 .38 3.27 .39 3.27 .37 7.9 48.6 Mean (individualistic countries) 3.24 .35 3.26 .37 3.23 .31 5.5 45.2 Mean (non-individualistic countries) 3.29 .40 3.28 .41 3.29 .39 9.7 51.3

Note. Imputed individualism scores are set in italics. †Low and medium educated significantly higher mean acquiescent scores compared with high educated (p < .05). {High educated significantly higher mean acquiescent scores compared with low educated (p < .05). }Low and medium educated significantly higher standard deviations of the acquiescent scores compared with high educated (p < .05).

74 B. Rammstedt et al.

DOI: 10.1002/per

potential method effects, we applied both the subtraction method and the regression-based approach to the total and the subsamples of one country, namely the USA. Results were highly similar across both methods.

Factor analyses To avoid variations in results due to methodological differences, we kept our method aligned with that applied in previous studies (Rammstedt et al., 2010; 2011) and with the assumptions of the Big Five model. PCAs with a forced extraction of five factors were conducted. All factor solutions were subsequently rotated to an optimal fit with (i) the ‘simple structure’ criterion (using Varimax rotation) and (ii) an idealized 10-item 5-dimensional Big Five factor structure (using Procrustes rotation as advocated by McCrae, Zonderman, Costa, Bond, & Paunonen, 1996 and Allik and McCrae, 2004).

Congruence of the factor structures The similarity of two 5-dimensional factor solutions, X and Y, can be assessed by computing the congruence coefficient c. To find the value of c which indicates a level of congruence that is significantly greater than what one can expect for random configurations X, Rammstedt et al. (2010) provided simulation norms. They reported that for the case studied in this paper (10 vectors, five dimensions, with Y the idealized Big Five pattern with +1, 0, and �1 loadings), a coefficient greater than .78 is the critical benchmark value of the random congruence. For random X, congruence of >.78 can be expected in less than 1% of the cases. Therefore, such similarity indices are considered statistically significant. However, these statistical norms are only weak benchmarks, because, as Lorenzo-Seva, & Ten Berge (2006) have shown, ‘seasoned factor analysts’ are likely to interpret X and Y as ‘fairly’ similar only if the congruence coefficient c is at least .85.

The hypothesized better matches of the empirical and the ideal factor structures are thus empirically testable using the following three criteria:

a. Congruence coefficients increase after controlling for acquiescence.

b. After controlling for acquiescence, congruence coefficients exceed the 99% norms for statistically significant congru- ence (c > .78).

c. After controlling for acquiescence, congruence coefficients should reach or exceed the .85 criterion proposed by Lorenzo-Seva, & Ten Berge (2006) if the factor structures are to be interpreted as at least ‘fairly’ similar.

RESULTS

Differences in acquiescent response bias (H1)

To test H1, means and standard deviations of the acquiescence scores for each country were computed. As we assume that the tendency towards acquiescence should be generally more pronounced and more variable in low-educated and medium- educated groups than in high-educated groups, Table 2 shows

the mean acquiescence scores and their variability for each country’s total sample (columns 3 and 4) and for each educational group (columns 5 and 6, and columns 7 and 8, respectively). Overall, the various countries differ in the extent and direction of the hypothesized educational effect. In seven of the 18 countries, the mean acquiescence scores of low-educated and medium-educated respondents were significantly higher than those of high- educated respondents. In six countries, the effect was not significant, and in the remaining five countries, an inverse effect was found, with high-educated respondents scoring higher on acquiescence than low-educated and medium-educated respondents. Concerning the variability of acquiescence scores, expected differences were found for eight of the 18 countries. In these countries, the standard deviations for low-educated and medium-educated groups were higher compared with those of the high-educated groups. For the remaining countries, no significant differences were observed. Notably, no effects contrary to expectations were substantial. In sum, these results do not support our hypothesis H1 for all countries. About half of the countries showed the predicted effects with regard to educational differences in mean and variability; for the other half, no such effects were found.

Factor structure of the raw scores and the scores corrected for acquiescence (H2)

According to H2, we assume that variability in the general tendency towards acquiescent responding blurs the Big Five factor structures in all countries. Controlling for acquiescence variance should universally improve the congruence of the factor structures with an ideal Big Five pattern. To test this hypothesis, we conducted two sets of factor analyses on the basis of the 10 BFI-10 variables for each of the 18 countries separately. One set was based on the raw data, and one on the data corrected for acquiescence.

In order to keep the results presented here to a reasonable size, we will as an example report detailed results for one country only, namely the USA. For all remaining countries, summarized results will be reported.

The top half of Table 3 shows the Procrustes-rotated4

factor-loading matrix for the 10 raw BFI-10 items for the U.S. sample. In this table, pairs of items measuring the same factor are re-ordered and listed in adjacent rows. In addition, the factors are ordered (from left to right) according to their loadings on the items. Results for the total sample are displayed on the left. The explained variance is 66%. As can be seen from the loading matrix, the Big Five factor solution is distinctly blurred and does not match the criterion of simple structure. Every item — with the exception of the two Emotional Stability items — has substantial loadings on non-corresponding factors. The ambiguity of the structure is also reflected in the factor congruence coefficients that measure the fit of the empirical factor loadings with the ideal

4Results based on Varimax rotation are highly similar to those of the Pro- crustes rotation for this and for all following factor analyses for the U.S. sample.

Correcting Big Five measurements for acquiescence 75

DOI: 10.1002/per

loadings (after optimal reflections and permutations of the factors). Following the benchmarks proposed by Lorenzo- Seva, & Ten Berge (2006) congruence coefficients of .85 or higher (set in bold in Table 4) can be interpreted by ‘seasoned factor analysts’ as an indication that the two structures are ‘fairly’ similar. For the U.S. data, the congruence coefficients vary between .66 and .96 for the five factors, with a total matrix fit of .86, which just meets the benchmark value.

To test whether controlling for acquiescence variance does indeed increase the clarity of the Big Five factor structures (i.e. H2), we repeated the factor analyses for each country by using the mean-corrected scores described in the Method section.

The lower half of Table 3 shows the detailed factor analytical results based on the mean-corrected data for our example country, the USA. The explained variance is 71%. The factor-loading matrix based on the scores corrected for acquiescence reflects the Big Five with textbook-like clarity: All items load substantially and highest on their corresponding factors, and none of the secondary loadings exceed .35. The good fit of the factor structure is also reflected in the congruence coefficients, which range between .92 and .96 for the five factors, with a total matrix congruence of .95. Therefore, correcting for acquiescence yields a markedly better factor structure in the U.S. sample.

Summarized factor analytical results for all 18 countries are reported in Table 4 in the form of congruence coefficients for the full matrix — for Varimax and Procrustes rotation separately. The countries in the table are sorted in descending order according to their level of individualism. Columns 2

and 10 show the congruence scores for the raw data, and columns 6 and 14 feature the scores for the data corrected for acquiescence. As can be seen from the table, the two methods of rotation yield similar results, even though congruence scores based on Procrustes rotation are necessarily always somewhat higher than those based on Varimax rotation.

In the case of the raw score data, the factorial fit is rather poor overall, which indicates that factor structures are blurred in all countries. The various countries differ in their mean congruence. However, a congruence of .85 is reached in only two of the 18 countries — the USA (in the case of the Procrustes rotation only) and Switzerland — which indicates theoretically unacceptable factor structures in at least 16 of the 18 countries.

Whether the factorial match increases when acquiescence is controlled for is tested by the three criteria formulated in the Method section. Our first criterion is a mere increase in congruence of the mean-corrected data compared with the raw data. This criterion is fully met. In all countries, and under both methods of rotation, congruence coefficients based on the corrected data are higher than those based on the non-corrected data.

According to the second criterion, congruence coefficients based on the corrected data should meet the benchmark of >.78 that allows rejecting random congruence (cf. Rammstedt et al., 2010) more often than the coefficients based on the uncorrected data. This was indeed the case. Based on the uncorrected data, a congruence significantly deviating from chance was reached in only six countries under

Table 3. Results for the USA: Procrustes-rotated factor structures of the raw and mean-corrected BFI-10 items for the total sample and for each educational group

I see myself as someone who . . .

Total Low and medium education High education

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

Raw scores . . . is reserved .89 .19 �.02 .04 .06 .76 �.25 .37 .06 .00 .88 .20 �.03 �.03 .04 . . . is outgoing, sociable �.67 .37 �.01 .19 .21 �.72 .02 .25 .22 .12 �.74 .30 .06 .15 .17 . . . is generally trusting �.16 .78 .10 �.10 .19 �.37 .20 .66 �.19 .03 �.15 .79 .10 �.08 .20 . . . tends to find fault with others .03 �.30 �.38 �.28 .36 �.10 �.73 �.06 �.23 .12 �.05 �.48 �.21 �.32 .26 . . . does a thorough job �.03 .47 .56 .01 .24 �.05 .31 .66 .06 .18 �.09 .42 .65 �.05 .15 . . . tends to be lazy �.01 �.01 �.81 �.14 .02 �.18 �.42 �.39 �.23 .01 .01 .11 �.87 �.16 .02 . . . is relaxed, handles stress well .04 .21 .00 .80 .13 �.04 �.05 .29 .74 .08 .01 .20 �.03 .81 .11 . . . gets nervous easily .19 .02 �.14 �.79 �.04 .12 �.09 .00 �.80 �.07 .18 �.04 �.15 �.79 �.02 . . . has an active imagination �.05 .21 �.11 .17 .75 �.17 �.37 .38 .19 .55 �.14 .22 �.08 .07 .73 . . . has few artistic interests .10 .38 �.32 .01 �.65 �.05 �.28 .20 .03 �.89 �.01 .28 �.21 �.07 �.73 Factor congruence .96 .66 .86 .94 .88 .91 .62 .61 .91 .94 .97 .77 .94 .94 .93 Scores corrected for acquiescence . . . is reserved .91 .04 .08 .05 �.01 .89 .03 .07 .10 .02 .92 .04 .06 .02 �.02 . . . is outgoing, sociable �.75 .16 .15 .22 .12 �.72 .16 .09 .25 .11 �.78 .15 .17 .20 .12 . . . is generally trusting �.23 .76 .20 �.11 .13 �.27 .71 .26 �.15 .07 �.18 .77 .18 �.08 .18 . . . tends to find fault with others �.11 �.80 �.01 �.28 .12 �.13 �.79 .00 �.32 .09 �.08 �.82 �.01 �.24 .14 . . . does a thorough job �.10 .23 .71 .01 .14 �.04 .33 .62 .00 .16 �.13 .19 .76 .01 .11 . . . tends to be lazy �.03 .02 �.87 �.17 .04 �.02 .07 �.88 �.20 .06 �.02 .01 �.87 �.15 .01 . . . is relaxed, handles stress well .01 .08 .08 .84 .06 .02 .12 .08 .81 .09 �.01 .04 .07 .87 .06 . . . gets nervous easily .18 �.08 �.10 �.80 �.09 .16 �.05 �.12 �.77 �.10 .18 �.13 �.09 �.81 �.06 . . . has an active imagination �.13 �.07 .06 .23 .70 �.10 �.16 .10 .28 .69 �.18 .00 .05 .15 .70 . . . has few artistic interests �.01 �.08 �.03 .07 �.90 .00 �.14 .00 .09 �.91 �.03 �.03 �.05 .04 �.90 Factor congruence .95 .96 .96 .92 .96 .95 .92 .94 .89 .96 .96 .97 .97 .95 .96

Note. Loadings ≥ .35 are set in bold.

76 B. Rammstedt et al.

DOI: 10.1002/per

T ab le

4 . T o ta l m at ri x fa ct o r co n g ru en ce

co ef fi ci en ts fo r th e ra w

d at a an d th e d at a co rr ec te d fo r ac q u ie sc en ce

b ia s fo r th e 1 8 co u n tr ie s (t o ta l sa m p le

an d ea ch

ed u ca ti o n al

g ro u p )

C o u n tr y

(i n d iv id u al is m )

V ar im

ax -r o ta te d st ru ct u re s

P ro cr u st es -r o ta te d st ru ct u re s

R aw

d at a

C o rr ec te d d at a

R aw

d at a

C o rr ec te d d at a

T o ta l

L o w /

m ed iu m

ed u ca ti o n

H ig h

ed u ca ti o n D if fe re n ce

T o ta l

L o w /

m ed iu m

ed u ca ti o n

H ig h

ed u ca ti o n D if fe re n ce

T o ta l

L o w /

m ed iu m

ed u ca ti o n

H ig h

ed u ca ti o n D if fe re n ce

T o ta l

L o w /

m ed iu m

ed u ca ti o n

H ig h

ed u ca ti o n D if fe re n ce

U S A

(3 )

.8 3

.7 6

.9 0

.1 4

.9 5

.9 3

.9 6

.0 3

.8 6

.8 0

.9 1

.1 1

.9 5

.9 3

.9 6

.0 3

G er m an y (3 )

.8 3

.8 2

.0 0

.9 5

.9 4

�. 0 1

.8 4

.8 3

.8 2

�. 0 1

.9 5

.0 0

Ir el an d (3 )

.7 3

.6 3

.7 5

.1 2

.8 9

.7 2

.9 1

.1 9

.7 9

.7 5

.8 3

.0 8

.9 1

.8 4

.9 1

.0 7

N ew

Z ea la n d (3 )

.7 5

.6 8

.7 9

.1 1

.9 4

.9 0

.9 5

.0 5

.7 9

.7 5

.8 1

.0 6

.9 4

.9 2

.9 5

.0 3

F ra n ce

(3 )

.7 9

.7 0

.8 3

.1 3

.8 7

.8 5

.8 7

.0 2

.8 2

.7 9

.8 4

.0 5

.8 8

.8 6

.8 8

.0 2

D en m ar k (3 )

.8 1

.7 9

.8 1

.0 2

.9 3

.9 2

.9 4

.0 2

.8 2

.8 0

.8 3

.0 3

.9 4

.9 3

.9 4

.0 1

S w it ze rl an d (3 )

.8 7

.8 5

.8 7

.0 2

.9 1

.9 3

.9 0

�. 0 3

.8 7

.8 5

.8 8

.0 3

.9 2

.9 3

.9 2

�. 0 1

B el g iu m

(F la n d er s; 3 )

.7 8

.7 3

.8 7

.1 4

.9 5

.9 4

.9 5

.0 1

.7 9

.8 8

.0 9

.9 5

.9 4

.9 5

.0 1

L at v ia

(2 )

.6 5

.6 3

.7 0

.0 7

.7 0

.7 4

.0 4

.6 8

.6 9

.7 9

.1 0

.7 4

.7 7

.0 3

R u ss ia

(2 )

.6 3

.6 2

.6 4

.0 2

.8 9

.8 5

.8 8

.0 3

.7 2

.7 0

.7 5

.0 5

.9 1

.8 9

.9 0

.0 1

C ze ch

R ep u b li c (2 )

.5 1

.5 0

�. 0 1

.6 9

.6 1

�. 0 8

.6 2

.6 1

�. 0 1

.7 3

.7 0

�. 0 3

Is ra el

(2 )

.6 0

.6 5

.6 8

.0 3

.8 3

.7 3

.7 0

�. 0 3

.7 5

.7 4

.7 3

�. 0 1

.8 7

.8 1

.8 3

.0 2

Ja p an

(2 )

.6 7

.6 6

.6 8

.0 2

.8 3

.8 0

.6 7

�. 1 3

.7 7

.7 6

.7 7

.0 1

.8 6

.8 4

.8 8

.0 4

P h il ip p in es

(1 )

.4 8

.4 4

.5 2

.0 8

.6 3

.5 6

.6 8

.1 2

.5 8

.5 5

.6 3

.0 8

.7 0

.6 9

.7 2

.0 3

M ex ic o (1 )

.5 4

0 .8 3

.8 5

.5 4

�. 3 1

.6 7

.6 6

�. 0 1

.8 7

.7 6

�. 1 1

T ai w an

(1 )

.7 1

.7 0

.7 3

.0 3

.9 2

.8 7

.7 1

�. 1 6

.7 8

.7 7

.8 0

.0 3

.9 2

.8 9

.9 2

.0 3

S o u th

K o re a (1 )

.7 9

.6 6

.8 2

.1 6

.8 4

.8 3

.7 9

�. 0 4

.7 9

.7 6

.8 2

.0 6

.8 4

.8 8

.0 4

D o m in ic an

R ep u b li c (1 )

.5 6

.5 1

.5 5

.0 4

.6 1

.6 0

.5 6

�. 0 4

.6 9

.6 6

.7 2

.0 6

.7 4

.7 2

.8 3

.1 1

T o ta l m ea n

.7 0

.6 6

.7 2

.0 6

.8 4

.8 1

.7 9

�. 0 2

.7 6

.7 4

.7 8

.0 4

.8 7

.8 5

.8 8

.0 1

M ea n

(i n d iv id u al is ti c

co u n tr ie s)

.8 0

.7 5

.8 3

.0 8

.9 2

.8 9

.9 3

.0 4

.8 2

.8 0

.8 5

.0 5

.9 3

.9 1

.9 3

.0 2

M ea n

(n o n -i n d iv id u al is ti c

co u n tr ie s)

.6 1

.5 9

.6 4

.0 5

.7 8

.7 5

.6 9

�. 0 6

.7 1

.6 9

.7 3

.0 3

.8 2

.8 0

.8 2

.0 1

N o te . Im

p ut ed

in di v id ua li sm

sc o re s ar e se t in

it al ic s. F ac to r co n gr u en ce

co ef fi ci en ts eq u al

to o r ex ce ed in g .8 5 ar e se t in

b ol d .

Correcting Big Five measurements for acquiescence 77

DOI: 10.1002/per

Varimax rotation and nine countries under Procrustes rotation; based on the corrected data, the criterion was satisfied in 14 of the 18 countries.

Our third criterion requires that the congruence coeffi- cients should be even higher (≥ .85) before the factor structures can be interpreted as a theoretically good fit. Based on the uncorrected data, this criterion was met in only one country under Varimax and in two countries under Procrustes rotation. Based on the data corrected for acquiescence, a good fit was reached by 10 of the 18 countries using Varimax and by 13 countries using Procrustes rotation. Therefore, H2, which hypothesizes better-fitting factor structures after correcting for acquiescent response bias, is clearly supported in terms of all three criteria.

Educational differences in the factor structures (H3 and H4)

In order to test H3 — that the factor structures in the low- educated and medium-educated samples are more severely blurred than those in the high-educated samples — factor analyses based on the raw score data were repeated separately for the two educational groups. For our sample country, the USA, Procrustes-rotated factor-loading matrices of the raw data are reported in the top half of Table 3. The results clearly support our hypothesis — at least for the USA, where the Big Five emerged in an unbiased way only in the high-educated samples. In the low-educated and medium-educated samples, the factor structure is distinctly blurred. Differences in the factor structures are also reflected in the congruence coefficients of the factors. For the low-educated and medium-educated sample, these coefficients vary between .61 and .94, with a congruence of the total matrix of .80. For the high-educated sample, by contrast, congruence ranges between .77 and .97, with a total matrix fit of .91.

Summarized factor analytical results for the raw data of all 18 countries for each educational group are reported in Table 4 (columns 3 and 4, and 11 and 12). Once again, results are given in terms of the full matrix congruence coef- ficients based both on Varimax and on Procrustes rotation.

As hypothesized, the congruence coefficients are lower in the low-educated and medium-educated group than in the high-educated group in 16 of the 18 countries under Varimax and in 14 countries under Procrustes rotation. In the two (Varimax) and four (Procrustes) remaining countries, respectively, no differences between the educational groups were observable. Hence, the results support — at least tentatively — the assumption of a more severely blurred factor structure in the low-educated and medium-educated group compared with the high-educated group (H3).

Our hypothesis H4 assumes that statistically controlling for acquiescence variance yields better-fitting Big Five patterns for all educational levels. This increase is expected to be more pronounced for those with a low or medium level of education. To test this hypothesis, we again conducted factor analyses for each educational group on the basis of the data corrected for acquiescence. For our sample country, the USA, factor-loading matrices of the corrected data are

reported in the bottom half of Table 3. To test for the better fit of the factor structures, we again used the three criteria formulated in H4: Congruence coefficients would (i) increase; (ii) exceed .78; and (iii) exceed .85. In the case of the USA, the results clearly met these three criteria. We found a higher increase in the factorial fit for the low-educated and medium-educated samples; for both educational groups, both benchmarks were met for each factor. In addition, the factor-loading matrices reflect a clear simple structure, unambiguously interpretable in terms of the Big Five. In both educational groups, all items load substantially and highest on their corresponding factors and none of the secondary loadings exceeds .35.

Results in terms of mean congruence coefficients for all 18 countries under both Varimax and Procrustes rotation are given in Table 4. Columns 7 and 15 display the results for the low-educated and medium-educated respondents, whereas columns 8 and 16 show the results for the high-educated group. For the low-educated and medium- educated respondents in all 18 countries, we found higher congruence coefficients under both methods of rotation after controlling for acquiescence. When comparing the high educated on the basis of both raw and corrected data, this was true for 14 (Varimax) and 17 (Procrustes) of the 18 countries, respectively. The increase in congruence was slightly higher on average for the low-educated and medium-educated respondents (.15 and .11 for Varimax and Procrustes, respectively) compared with the high-educated group (.07 and .09 for Varimax and Procrustes, respectively).

For the other two criteria, the picture was even clearer: In the low-educated and medium-educated samples, based on the raw data, only three (Varimax) and six (Procrustes) countries, respectively, met the .78 criterion and only one (Switzerland) just met the .85 criterion. After controlling for acquiescence, this ratio increased to 12 (Varimax) and 14 (Procrustes) countries, respectively, for the .78 benchmark and to 10 countries for the .85 criterion. In the high-educated samples — even based on the raw data — 8 (Varimax) and 11 (Procrustes) countries, respectively, met the .78 criterion and 3 fulfilled the .85 criterion. After controlling for acquiescence, 10 (Varimax) and 14 (Procrustes) countries, respectively, exceeded .78, whereas 9 (Varimax) and 12 (Procrustes) countries, respectively, exceeded .85.

In sum, all three criteria indicate an increase in the factorial fit after controlling for variance in acquiescence, and this increase is bigger for the low-educated and medium-educated respondents than for the high-educated respondents. Therefore, results indicate (i) a clear improvement of the factor structures; and (ii) diminished educational differences when acquiescence is controlled for.

Investigating differences between countries (H5 and H6)

Finally, we set out to test whether systematic differences among the countries can be found in the sense that individualistic countries show effects similar to those found for Germany in previous studies (H5 and H6).

The hypotheses are clearly supported by the data. The educational bias in acquiescence is more pronounced in the

78 B. Rammstedt et al.

DOI: 10.1002/per

individualistic cultures. For six of the eight individualistic countries, we found the hypothesized effects, with low- educated and medium-educated respondents scoring significantly higher than high-educated respondents. Similarly, we found for five of the eight countries that lower-educated and medium-educated respondents show a markedly higher variability in their acquiescence scores compared with high-educated respondents. In the remaining two, respectively three countries, the educational groups did not differ significantly. In the 10 non-individualistic cultures, however, only one country (Israel) showed both the mean level and the variability effect in the intended direction. No other country showed the hypothesized mean-level effects and only two other countries the assumed variability effects.

The fit of the factor structures also vary with the country’s classification on the individualism/collectivism dimension. In the individualistic countries, a very homoge- neous picture emerges. Even in the case of the raw score data, the congruence of the Big Five patterns is higher for these countries than for the non-individualistic ones [on average .80 (Varimax) and .82 (Procrustes), respectively, compared with .61 and .71, respectively]. This difference remained more or less constant after controlling for acquiescence (.92 and .93, respectively, compared with .78 and .82, respectively). In all eight individualistic countries — under both methods of rotation — and in both educational groups — with the excep- tion of Ireland for the low-educated and medium-educated respondents — congruence exceeds the criterion of .85 for an acceptable fit. In the 10 non-individualistic countries, by contrast, the results were much more heterogeneous. For some countries (Russia and Taiwan), we found good fitting results (both for the total sample and the two educational groups). For most of the countries, however, congruence did not reach the .85 criterion (for either the total sample or the two educational groups).

DISCUSSION

Previous research based on representative samples of the German adult population (Rammstedt & Kemper, 2011; Rammstedt et al., 2010) has found that interindividual differences in acquiescent responding may blur the Big Five factor structure. The aim of the present study was to investigate whether this effect is, in fact, universal and can thus be replicated in countries or cultures other than Germany. On the basis of the results of our own and other previous studies in this field, we formulated six hypotheses: We expected acquiescent responding to be more pronounced and more variable among low-educated and medium-educated respondents compared with high- educated respondents in all countries (H1). Furthermore, we hypothesized that individual differences in the tendency to acquiesce blur the Big Five factor structures and that correcting for this response bias leads to clearer Big Five factor structures in all countries (H2). As low- educated and medium-educated respondents are assumed to be more prone to acquiescent responding and within the group more variable, we hypothesized that the Big Five factor

structures of low-educated and medium-educated respondents would be relatively more blurred (H3). Moreover, we anticipated that correcting for acquiescence would yield better-fitting Big Five structures for both educational groups but that this increase would be more pronounced in the low-educated and medium-educated group (H4). On the basis of previous research indicating systematic cultural differences in the quality of Big Five solutions and in acquiescence, we expected that individualistic countries would display better-fitting factor structures overall; with regard to acquiescence, we anticipated relatively homogeneous results in the sense that the effects previously identified for Germany would be replicable in all individualistic countries (H5 and H6). We did not formulate any clear-cut hypotheses for the non-individualistic cultures. Rather, we assumed a more heterogeneous picture overall.

We tested our hypotheses on the basis of large data sets representative of the adult population of 18 countries from all over the world. The results of our analyses clearly support most of our hypotheses. We identified a general tendency towards acquiescent responding, which seems to have a biasing effect on the Big Five structures in all countries. Controlling for acquiescence markedly increased the fit of the factor structures in each of the 18 countries (H2).

The fit of the factor structures differed systematically across countries (cf. H5 and H6). As hypothesized, we found mark- edly higher and more homogeneous congruence coefficients for the individualistic countries, where all indices exceeded our statistical benchmark criterion, even based on the raw score analyses. Our findings are thus comparable to those reported by De Raad, Perugini, Hrebickova, and Szarota (1998), whose investigation of Big Five congruences in eight Western countries yielded coefficients between .77 and .85. However, as De Raad et al. investigated only student samples and, thus, samples comparable with our high-educated group, our find- ings even appear to outperform theirs.

For both individualistic and non-individualistic cultures, we found roughly the same increase in factorial fit after controlling for acquiescence variance so that, based on the corrected data, the factorial fit is also markedly higher in the individualistic countries. The match of the corrected data to the ideal Big Five factor structure in these eight countries is notable. In all countries, congruence coefficients exceeded the criterion of .85 suggested by Lorenzo-Seva, & Ten Berge (2006).

With regard to the assumed educational bias, the results are not as clear as those for the total samples. Even though we did not find a general educational bias in the sense that low-educated and medium-educated persons have a higher average level and variability in acquiescent responding (H1), we did find the predicted educational bias for all individualistic countries, as postulated by H5.

The hypothesized educational effects in the factorial fit of the Big Five structures (H3) are also supported by the data. In most countries, the Big Five patterns of the low-educated and medium-educated respondents showed a weaker fit compared with those of the high-educated respondents. Here again, we found some support for our cultural-difference hypothesis H5: On average, the educational bias based on the uncorrected

Correcting Big Five measurements for acquiescence 79

DOI: 10.1002/per

factorial structures was slightly higher in the individualistic cultures compared with the non-individualistic cultures.

Finally, the differential effectiveness of the correction for acquiescence (H4) is clearly supported by the data, with low-educated and medium-educated samples benefitting more from the correction than the samples comprising high-educated respondents. All three criteria defined a priori were met.

In sum, we found clear support for our hypotheses. A universal, culture-independent tendency towards individual differences in acquiescent responding that blurs the Big Five factor structures was identified. Therefore, our alternative hypothesis H5, which assumed such an effect primarily for the individualistic countries, can be regarded as too conser- vative. However, the postulated educational differences do not seem to be universally replicable. On the contrary — as postulated by H5 — they appear to vary with the country’s cultural values. For the individualistic countries, we clearly found the hypothesized educational bias, whereas in the case of the non-individualistic cultures, the pictures that emerged were less clear.

It might be argued that the distorting effects found in the present study must not necessarily solely be due to acquiescence but might be due to misunderstandings of item content that could be especially pronounced in the lower educated. When looking at the raw score data structure only, this was also our first impression, namely that personality questionnaire items are only understood by the higher educated. We were happy to be able to confute this assumption by showing that controlling for acquiescence diminishes the educational bias in the factor structures and yields perfectly fitting structures for both educational groups.

Alternatively, one might argue that the better-fitting factor structures in higher educated may stem from a more pronounced tendency for social desirable responding in the higher educated. The university-educated respondents might be more likely to endorse the items in a desirable direction, at least compared with less educated respondents who might be relatively naive about such impression management tactics. If this assumption holds, however, we should have found pronounced differences in scale means between high and low/medium-educated groups, which we did not (all effect sizes were below .3), but no differences in the mean response across all items as reflected in the acquiescence score, which we did find.

A clear limitation of the present study might be seen in the fact that only an extremely brief Big Five measure was assessed. Consequently, the question must be raised as to what extent the here-found results are generalizable across different Big Five questionnaires and therefore to Big Five assessment in general. Soto’s study (Soto et al., 2008) gives first support to this notion. He was able to identify the distorting impact of acquiescence on the Big Five factor structures also based on the full-scale BFI, however contrasting children to adults. On the basis of adult samples varying in their highest education, we could show in a recent study (Rammstedt, 2012) that our here- found results can be clearly replicated also based on a

full-scale Big Five measure: Based on the raw score data, the Big Five factor pattern could hardly be replicated in the less educated groups. After controlling for acquiescence, however, we found perfectly fitting Big Five factor structures in all subsamples. Thus, it can be assumed that the here-found effects generalize also across different and more detailed Big Five questionnaires.

Our main goal was to investigate whether the distorting impact of acquiescent responding on the Big Five factor structure observed for samples representative of the German population (Rammstedt & Kemper, 2011; Rammstedt et al., 2010) generalizes across countries. Results of the present study clearly support this assumption. We found a general tendency towards acquiescent responding in all countries, a distorting impact on the Big Five factor structure, and an improvement of fit to the Big Five model after statistically controlling for acquiescence variance.

REFERENCES

Allik, J., & McCrae, R. R. (2004). Escapable conclusions: Toomela (2003) and the universality of trait structure. Journal of Personality and Social Psychology, 87, 261–265.

Ayidiya, S. A., & McClendon, M. J. (1990). Response effects in mail surveys. Public Opinion Quarterly, 54, 229–247.

Benet-Martínez, V., & John, O. P. (1998). Los Cinco Grandes across cultures and ethnic groups: Multitrait multimethod analyses of the Big Five in Spanish and English. Journal of Personality and Social Psychology, 75, 729–750.

Church, T. A., & Katigbak, M. S. (2000). Trait psychology in the Philippines. American Behavioral Scientist, 44, 73–94.

Clarke I., III. (2001). Extreme response style in cross-cultural research. International Marketing Review, 18, 301–324.

Clemans, W. V. (1966). An analytical and empirical examination of some properties of ipsative measures. Richmond: The William Byrd Press.

Cunningham, W., Cunningham, I. C. M., & Green, R. T. (1977). The ipsative process to reduce response set bias. Public Opinion Quarterly, 41, 379–394.

De Raad, B. (2000). The Big Five personality factors: The psycholexical approach to personality. Ashland, Ohio: Hogrefe & Huber Publishers.

De Raad, B., Perugini, M., Hrebickova, M., & Szarota, P. (1998). Lingua Franca of personality: Taxonomies and structures based on the psycholexical approach. Journal of Cross-Cultural Psychology, 29, 212–232.

Diamantopoulos, A., Reynolds, N. L., & Simintiras, A. C. (2006). The impact of response styles on the stability of cross-national comparisons. Journal of Business Research, 59, 925–935.

Fischer, R. (2004). Standardization to account for cross-cultural response bias: A classification of score adjustment procedures and review of research in JCCP. Journal of Cross-Cultural Psychology, 35, 263–282.

Flavell, J. H., Miller, P. H., & Miller, S. A. (1993). Cognitive developments. Engelwood Cliffs, New Jersey: Prentice-Hall.

Goldberg, L. R. (1963). A model of item ambiguity in personality assessment. Educational and Psychological Measurement, 23, 467–492.

Goldberg, L. R. (1990). An alternative “description of personality”: The Big-Five factor structure. Journal of Personality and Social Psychology, 59, 1216–1229.

Greenfield, P. (1999). Three approaches to the psychology of culture: Where do they come from? Where can they go? Presented at 3rd Conference of the Asian Association of Social Psychology, Taiwan.

80 B. Rammstedt et al.

DOI: 10.1002/per

Guanzon-Lapena, M. A., Church, A. T., Carlota, A. J., & Katigbak, M. S. (1998). Indigenous personality measures: Philippine examples. Journal of Cross-Cultural Psychology, 29, 249–270.

Haller, M., Jowell, R., & Smith, T. W. (Eds.). (2009). Charting the globe: The International Social Survey Programme, 1984–2009. London: Routledge.

Hofstede, G. (2001). Culture’s consequences, comparing values, behaviors, institutions, and organizations across nations. Thousand Oaks, CA: Sage Publications.

Hofstede, G., & Hofstede, G.-J. (2004). Cultures and organizations: Software of the mind. New York: McGraw-Hill.

Jackson, D. N., & Messick, S. (1958). Content and style in personality assessment. Psychological Bulletin, 55, 243–252.

Javeline, D. (1999). Response effects in polite cultures: A test of acquiescence in Kazakhstan. Public Opinion Quarterly, 63, 1–28.

John, O. P., Donahue, E. M., & Kentle, R. L. (1991). The “Big Five” Inventory — Versions 4a and 54. Berkeley: University of California, Berkeley, Institute of Personality and Social Research.

John, O. P., Naumann, L. P., & Soto, C. J. (2008). Paradigm shift to the integrative Big-Five trait taxonomy: History, measurement, and conceptual issues. In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of personality: Theory and research (pp. 114–158). New York: Guilford Press.

Johnson, T., Kulesa, P., Cho, Y. I., & Shavitt, S. (2005). The rela- tionship between culture and response styles: Evidence from 19 countries. Journal of Cross-Cultural Psychology, 36, 264–277.

Krosnick, J. A., Narayan, S., & Smith, W. R. (1996). Satisficing in surveys: Initial evidence. New Directions for Evaluation, 70, 29–44.

Lang, F. R., Lüdtke, O., & Asendorpf, J. (2001). Testgüte und psychometrische Äquivalenz der deutschen Version des Big Five Inventory (BFI) bei jungen, mittelalten und alten Erwachsenen. [Validity and psychometric equivalence of the German version of the Big Five Inventory in young, middle-aged and old adults]. Diagnostica, 47, 111–112.

Lorenzo-Seva, U., & Ten Berge, J. M. F. (2006). Tucker’s con- gruence coefficient as a meaningful index of factor similarity. Methodology, 2, 57–64.

McClendon, M. J. (1991). Acquiescence: Tests of the cognitive limitations and question ambiguity hypotheses. Journal of Official Statistics, 7, 153–166.

McCrae, R. R., Zonderman, A. B., Costa, P. T., Jr., Bond, M. H., & Paunonen, S. V. (1996). Evaluating replicability of factors in the Revised NEO personality inventory: Confirmatory factor analysis versus Procrustes rotation. Journal of Personality and Social Psychology, 70, 552–566.

Meisenberg, G., & Williams, A. (2008). Are acquiescent and extreme response styles related to low intelligence and educa- tion? Personality and Individual Differences, 44, 1539–1550.

Mirowsky, J., & Ross, C. E. (1991). Eliminating defense and agreement bias from measures of the sense of control: A 2 � 2 index. Social Psychology Quarterly, 54, 127–145.

Narayan, S., & Krosnick, J. A. (1996). Education moderates some response effects in attitude measurement. Public Opinion Quarterly, 60, 58–88.

Ostendorf, F. (1990). Sprache und Persönlichkeitsstruktur. Zur Validität des Fünf-Faktoren-Modells der Persönlichkeit. [Language and personality structure. On the structural validity of the Five Factor Model of personality]. Regensburg: Roderer.

Rammstedt, B. (1997). Die deutsche Version des Big Five Inventories (BFI): Übersetzung und Validierung eines Fragebogens zur Erfassung des Fünf-Faktoren-Modells der Persönlichkeit [The German version of the Big Five Inventory (BFI): Translation and validation of a questionnaire for the measurement of the Five Factor model of personality]. Unpublished thesis, University of Bielefeld, Bielefeld, Germany.

Rammstedt, B. (2012). The effects of acquiescence on the Big Five personality structure. Manuscript in preparation.

Rammstedt, B., & John, O. P. (2007). Measuring personality in one minute or less: A 10-item short version of the Big Five Inventory in English and German. Journal of Research in Personality, 41, 203–212.

Rammstedt, B., & Kemper, C. J. (2011). Measurement equivalence of the Big Five: Shedding further light on potential causes of the educational bias. Journal of Research in Personality, 45, 121–125.

Rammstedt, B., & Krebs, D. (2007). Does response scale format affect the answering of personality scales? Assessing the Big Five dimensions of personality with different response scales in a dependent sample. European Journal of Psychological Assessment, 23, 32–38.

Rammstedt, B., Goldberg, L. R., & Borg, I. (2010). The mea- surement equivalence of Big-Five factor markers for persons with different levels of education. Journal of Research in Personality, 44, 53–61.

Scholz, E., Harkness, J., & Faass, T. (2008): ISSP study monitoring 2004, report to the ISSP General Assembly on monitoring work undertaken for the ISSP. GESIS Technical Report. Mannheim: GESIS.

Soto, C. J., John, O. P., Gosling, S. D., & Potter, J. (2008). The developmental psychometrics of Big-Five self-reports: Acqui- escence, factor structure, coherence, and differentiation from ages 10 to 20. Journal of Personality and Social Psychology, 94, 718–737.

Ten Berge, J. M. F. (1999). A legitimate case of component analysis of ipsative measures, and partialling the mean as an alternative to ipsatization. Multivariate Behavioral Research, 34, 89–102.

Toomela, A. (2000). Stages of mental development: Where to look? Trames: A Journal of the Humanities in Social Sciences, 4, 21–52.

Toomela, A. (2003a). Relationship between personality structure, structure of word meaning, and cognitive ability: A study of cultural mechanisms of personality. Journal of Personality and Social Psychology, 85, 723–735.

Toomela, A. (2003b). Development of symbol meaning and the emergence of the semiotically mediated mind. In A. Toomela (Ed.), Cultural guidance in the development of the human mind (pp. 163–209). Westport, CA: Aplex Publishing.

Winkler, J. D., Kanouse, D. E., & Ware, J. E. (1982). Controlling for acquiescence response set in scale development. Journal of Applied Psychology, 67, 555–561.

Correcting Big Five measurements for acquiescence 81

DOI: 10.1002/per

not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written

permission. However, users may print, download, or email articles for individual use.