People tend to demonstrate substantial differences in how they respond to a typical survey item, and the array of these response styles might suggest certain important biological, physiological, and cultural characteristics of these individuals (Jackson & Messick, Reference Jackson and Messick1958). When a person consistently agrees to survey items regardless of their content, for example, their behaviors may demonstrate the manifestation of certain substantive traits. A traditional assumption concerning response bias is that it constitutes transient characteristics caused entirely by situational irregularities (e.g., Rorer, Reference Rorer1965). However, recent research in this area consistently suggests that response styles can be temporally stable (e.g., Alessandri et al., Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010; Couch & Keniston, Reference Couch and Keniston1960; Marsh et al., Reference Marsh, Scalas and Nagengast2010; Motl & DiStefano, Reference Motl and DiStefano2002; Quilty et al., Reference Quilty, Oakman and Risko2006). Our current research contributes to this area of study by examining the heritability of two of the most important response styles: acquiescence bias and item keying effect. If both acquiescence bias and item keying effect are heritable, then both may represent some substantive biological characteristics that are worthy of investigations on its own. Answers to the nature versus nurture question with regard to response styles will provide opportunities for future research studies on their etiology.
Acquiescence Bias and Item Keying Effect
Cronbach (Reference Cronbach1942) was one of the first researchers to popularize the issue of acquiescence bias (see also Block, Reference Block1965; Lentz, Reference Lentz1938; Lorge, Reference Lorge1937). When administering true–false tests with a group of students, Cronbach observed that some test takers were more likely to choose ‘true’ as an answer to an item compared to other test takers. Acquiescence bias refers to a participant's tendency to agree or disagree with survey items regardless of the items’ actual content. Cronbach's (Reference Cronbach1941, Reference Cronbach1942) seminal discovery on this response style has spawned decades of research in this area, including but not limited to the nature of acquiescence bias (Heal & Sigelman, Reference Heal and Sigelman1995), its stability within a survey and across time (Billiet & McClendon, Reference Billiet and McClendon2000; Weijters et al., Reference Weijters, Geuens and Schillewaert2010), its prevalence (Messick & Jackson, Reference Messick and Jackson1961), and its cross-cultural manifestations (Grimm & Church, Reference Grimm and Church1999; Kam et al., Reference Kam, Zhou, Zhang and Ho2012). In one of the early studies on acquiescence bias, Jackson and Messick (Reference Jackson and Messick1958) discovered that many participants agree with logically contradictory statements (i.e., regular- vs. reverse-keyed items; see also Messick & Jackson, Reference Messick and Jackson1961, for similar findings). Later researchers often regarded acquiescence bias as an error caused by cognitive impairment, low verbal ability, poor education, low social status, and cognitive aging (e.g., Ayidiya & McClendon, Reference Ayidiya and McClendon1990; Krosnick et al., Reference Krosnick, Narayan, Smith, Braverman and Slater1996; Lenski & Leggett, Reference Lenski and Leggett1960; Meisenberg & Williams, Reference Meisenberg and Williams2008; Messick & Frederiksen, Reference Messick and Frederiksen1958, Reference Messick and Frederiksen1961; Narayan & Krosnick, Reference Narayan and Krosnick1996; Rammstedt et al., Reference Rammstedt, Goldberg and Borg2010; Sigelman et al., Reference Sigelman, Budd, Spanhel and Schoenrock1981). These findings subsequently suggest that acquiescence bias was an error to be controlled for.
To mitigate the effect of acquiescence bias, researchers often include reverse-keyed items within questionnaires (e.g., Rammstedt, et al., Reference Rammstedt, Goldberg and Borg2010). Regular-keyed items measure the presence of a construct whereas reverse-keyed items measure the absence of a construct. When participants agree to both regular-keyed items and reverse-keyed items simultaneously, the acquiescence bias in regular-keyed and reverse-keyed items balances out. Ironically, attempts to control for acquiescence bias often unintentionally engender the second type of response style — item keying effect (Kam & Meyer, Reference Kam and Meyer2012; Marsh, Reference Marsh1996; Tomás & Oliver, Reference Tomás and Oliver1999).
According to Kam and Meyer (Reference Kam and Meyer2012), regular- and reverse-keyed items usually differ in the valence or the favorability of their content (Peabody, Reference Peabody1967). For instance, for the construct conscientiousness, regular-keyed items (e.g., ‘I am careful’) are more favorable in content compared to reverse-keyed items (e.g., ‘I am careless’). Participants may answer a questionnaire item based on the valence of the item in addition to its content. Item keying effect thus refers to participants’ differential responding styles between regular- and reverse-keyed items (Kam & Meyer, Reference Kam and Meyer2012). In a recent study, Biderman et al. (Reference Biderman, Nguyen, Cunningham and Ghorbani2011) found the existence of a small item keying effect even in Big Five personality measures. Given that the Big Five personality measure is widely used, the findings speak about the prevalence of the item keying effect in a great deal of psychological research.
Stability Nature of the Two Response Styles
Research has shown that both acquiescence bias and item keying effect may represent certain stable characteristics. Weijters et al. (Reference Weijters, Geuens and Schillewaert2010) recently modeled acquiescence bias with structural equation modeling (SEM). These authors split a survey into five successive sets and discovered that acquiescence bias consistently and equally existed in each set. Particularly, all observed indicators of acquiescence bias had identical loadings on the same latent factor in their SEM investigation, demonstrating that acquiescence bias is consistent within a survey. In another study, Billiet and McClendon (Reference Billiet and McClendon2000) found that the acquiescence latent factor in an SEM model correlates strongly with another operationalization of acquiescence bias (‘sum of agreements’ on survey items; r = 0.90), thus supporting the convergent validity of modeling acquiescence bias with SEM. Based on Weijters et al.'s (2010) research, the current investigation will assume the magnitude of acquiescence bias to be identical among different items within the same survey. Finally, researchers have discovered that acquiescence bias from the same survey correlated at around 0.6 over a span of four years (Bachman & O'Malley, Reference Bachman and O'Malley1984; Billiet & Davidov, Reference Billiet and Davidov2008) and over four 1-year periods (Marsh et al., Reference Marsh, Scalas and Nagengast2010; Motl & DiStefano, Reference Motl and DiStefano2002).
In addition to temporal stability, there is some evidence that item keying response style actually represents a heritable quality. Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) extracted response style factors associated with regular- and reverse-keyed items in an optimism measure and compared these factors between monozygotic (MZ) and dizygotic (DZ) twins. These authors discovered that MZ twins have substantially higher item keying effect than DZ twins. This result indicates that genetics may play a role in causing the higher correlation of an item keying effect among MZ twins. In addition, these authors estimated the heritability of item keying effects to be 20% for optimism items but only 6% for pessimism items. The heritability estimate was statistically significant for optimism items but non-significant for pessimism items. The researchers did not find that shared environmental influences predicted the item keying effect. Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) thus concluded that the item keying effect was found heritable for optimism items that were also positively valenced, but not for pessimism items that were also negatively valenced.
The conclusions by Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010), however, can be challenged on several grounds. First, some researchers have argued that optimism and pessimism are two correlated but distinct constructs. Therefore, any item keying effect extracted from optimism items (regular-keyed items) would represent a construct of optimism but not pessimism. Critics may thus interpret Alessandri et al.'s (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) findings as the construct optimism being more heritable than the construct pessimism, rather than item keying effect being heritable. Second, Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) only examined one psychological measure in their study. Therefore, any interpretation of their findings might be confined only to the construct of optimism and their results may not be generalized to other constructs such as personality traits. Third, Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) studied only item keying response style. Their conclusion may differ substantially when acquiescence bias is also estimated simultaneously with item keying response effect. Our empirical investigations, detailed in the next section, will help build upon this area of research.
The Current Research
The current study aims to advance the research of Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) in several ways. First, past research usually estimated either acquiescence bias (e.g., Weijters et al., Reference Weijters, Geuens and Schillewaert2010) or item keying effect (e.g., Alessandri et al., Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010; Motl & DiStefano, Reference Motl and DiStefano2002) but seldom assessed them simultaneously within a measure. The current study will examine whether acquiescence bias and item keying effect can exist simultaneously. This question is important because the estimation of one response style can sometimes disappear after another response style is controlled for. Second, we included multiple constructs in our study because response styles found in only one particular measure, as in Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010), may simply represent idiosyncratic characteristics of that measure. The HEXACO personality traits are used in the current study because they have been employed across multiple areas of psychology and each personality trait is theoretically unidimensional. If an item keying effect is found in these multiple unidimensional personality measures, it is likely to represent response patterns rather than representation of an unintended construct in a particular measure. Third, to our knowledge, the heritability of acquiescence bias has never been assessed in previous research. Our research is theoretically important because it assesses potential biological determinants in acquiescence bias. Finally, we will examine whether response styles affect heritability estimates; for example, does controlling for acquiescence bias affect the heritability estimates in twin studies in general? We will address the last research question by comparing the heritability estimates of personality constructs before and after controlling for response styles.
Method
Participants
Participants were 795 MZ pairs (726 female and 69 male pairs) and 662 same-sex DZ pairs (616 female and 46 male pairs). The average age of the twins was 58.56 years (SD = 12.85), and ranged in age from 18 to 86 years. Participants were part of a larger study (see Veselka et al., Reference Veselka, Schermer, Petrides, Cherkas, Spector and Vernon2009) and were part of the Department of the TwinsUK registry based at the Department of Twin Research and Genetic Epidemiology (DTR), King's College, London, England.
Self-completion questionnaires were sent to these adult twins, who are all volunteers in the TwinsUK Adult Twin Registry (Spector & Williams, Reference Spector and Williams2006). All were ascertained from the general population and shown to be comparable to age-matched population singletons. These unselected MZ and DZ twins have been recruited since 1992 using twin registers and national media campaigns and have been used in a wide variety of studies (www.twinsuk.ac.uk). For historical reasons, the cohort is predominantly female: when the twin study started, its purpose was to study bone diseases such as osteoporosis and osteoarthritis and the effects of hormones at menopause. After 2004, the study opened up to include males, but no targeted recruitment of males has been done. The twins in the registry are not selected for any particular trait and they volunteer to take part in studies that cover a wide range of traits and common medical conditions (Andrew et al., Reference Andrew, Hart, Snieder, De Lange, Spector and MacGregor2001). The study was approved by the St Thomas’ Hospital research ethics committee, and all twins in the study provided informed consent.
Materials and Procedure
Participants completed the 60-item HEXACO Personality Inventory (HEXACO-60; Ashton & Lee, Reference Ashton and Lee2009) as part of a mailed survey package (see Veselka et al., Reference Veselka, Schermer, Petrides, Cherkas, Spector and Vernon2009). The HEXACO assesses six dimensions of personality: honesty–humility, emotionality, extraversion, agreeableness, conscientiousness, and openness-to-experience. Items are responded by using a 1 (strongly disagree) to 5 (strongly agree) Likert scale. The internal consistency coefficients of the scales are shown in Table 1.
Model Comparisons
Note: The final best model based on Bentler–Satorra (2001) scaled difference χ2 tests is shown in bold. Because MLR was used for model estimators, we used the Satorra–Bentler (2001) test rather than the more common χ2 difference test. The former test (Satorra–Bentler test) is appropriate when MLR is the estimator but the latter test (common χ2 difference test) is appropriate when ML (maximum likelihood) is the estimator (Satorra & Bentler, Reference Satorra and Bentler2001).
TLI = Tucker-Lewis Index; CFI = Comparative Fit Index; RMSEA = Root Mean Square Error of Approximation; SRMR = Standardized Root Mean Square Residual.
Results
Multitrait–Multimethod Confirmatory Factor Analyses
The method effect in the HEXACO personality scale was evaluated using the correlated trait–uncorrelated method (CTUM) model within a framework of multitrait–multimethod (MTMM) confirmatory factor analysis (CFA). CTUM was chosen over other MTMM CFA models (e.g., correlated trait–correlated method; CTCM) because it does not overestimate the method variance, thereby providing a more conservative estimation of the method effect (Marsh, Reference Marsh1989; see also Marsh & Bailey, Reference Marsh and Bailey1991). Recent evidence by Marsh et al. (Reference Marsh, Scalas and Nagengast2010) supported the estimation accuracy of CTUM models. Following recommendations in the literature, we parceled personality items (e.g., Marsh et al., Reference Marsh, Hau, Balla and Grayson1998; Nasser & Wisenbaker, Reference Nasser and Wisenbaker2003; Yuan et al., Reference Yuan, Bentler and Kano1997). Parceled indicators are psychometrically more reliable than individual items. Items belonging to the same valence within a scale (e.g., positively valenced honesty items) were first factor-analyzed and then ordered according to their factor loadings. Based on a popular procedure, an item with high factor loadings is paired with another item with low factor loadings to form a parcel, so that the averaged loading becomes more balanced among the parcels (Yuan et al., Reference Yuan, Bentler and Kano1997). Two to three items were averaged to form a parcel in the model. Finally, we followed a statistical procedure suggested by Stapleton (Reference Stapleton2006) in order to control for the statistical dependency within each pair of twins in our model estimations. This procedure, using a robust maximum likelihood estimator (MLR), provides more accurate estimation of the model parameters when the assumptions of independent observations and normality are violated.
We compared five nested MTMM CFA models (see Figure 1). The first model, M baseline, is a common CFA model where observed personality item indicators load on their corresponding construct factors only. The second model, M positive, builds upon the baseline model by including a positive valence method factor. All positively valenced itemsFootnote 1 load on this method factor. The third model, M negative, includes a negative valence method factor and all negatively valenced items load on this method factor. The fourth model, M 2methods, includes both the positive and negative valence factors, whereas in the CTUM framework they are restricted to be orthogonally positioned to each other. If M 2methods fits better than the previous three models (M baseline, M positive, and M negative), it is likely due to the fact that participants are showing differential response styles to items of opposing valence. The fifth model, M 2m+r, is built upon M 2methods by including a random intercept (Maydeu-Olivares & Coffman, Reference Maydeu-Olivares and Coffman2006) in addition to the two valence factors. Acquiescence bias is modeled by this random intercept (Maydeu-Olivares & Coffman, Reference Maydeu-Olivares and Coffman2006).
MTMM models. Note: O = openness; C = conscientiousness; X = extraversion; A = agreeableness; E = emotionality; H = honesty–humility; p = positive valence factor; n = negative valence factor; acq = acquiescence bias. Each rectangle represents observed indicators that are measured by positively valenced items (e.g., Op) and negatively valenced items (e.g., On). Single arrows represent factor loadings and double arrows represent covariance among latent factors.
The five MTMM models were estimated with the Mplus 6.1 computer program (Muthén & Muthén, Reference Muthén and Muthén1998–2010; Table 1). Although M positive and M negative models fit better than M baseline, the M 2methods model showed better fit than any of these three models. This result replicates the finding by Biderman et al. (Reference Biderman, Nguyen, Cunningham and Ghorbani2011), suggesting that participants respond to positively valenced items and negatively valenced items differently. However, the best fitting model contains the positive and negative valence methods and a random intercept (M 2m+r) factor. Results suggest that both acquiescence bias and the two valence factors exist in participants’ responses. We also decomposed the variance explained by the personality factors, acquiescence bias, item keying response style, and residual variance. The results are as follows: 33.18% (personality factors), 5.73% (item keying), 2.11% (acquiescence bias), and 58.98% (residual variance). We then selected the best-fitting MTMM model (M 2m+r) and extracted the factor scores from the method factor(s) and the six HEXACO personality factors from the chosen model for heritability analyses.
Univariate Genetic Analyses
Analysis of the factor score residual covariance (controlling for sex and age), using standard SEM methods (Neale & Cardon, Reference Neale and Cardon1992), was performed (Neale et al., Reference Neale, Boker, Xie and Maes2006). In conducting univariate genetic analyses, a phenotypic score is expressed as a linear function of three factors: genetic (A), common environment (C), and specific environment (E). Following the finding of Sullivan and Eaves (Reference Sullivan and Eaves2002), a full ACE model was analyzed and not the reduced models. Table 2 presents the intraclass correlations for the MZ and DZ twins for each factor score as well as the estimated heritability (a 2), common environment (c 2), and specific environment (e 2) values from the standardized parameter estimates, and the 95% confidence interval (intervals not containing zero are deemed to be significant). All of the factor scores, except for the acquiescence bias, were found to have a significant genetic component, with values ranging from 0.24 for the positive valence factor to 0.54 for openness to experience. Common environmental effects were found to account for zero percent of the variance for personality factors and item keying factors. Unique environmental effects were significant for all of the factor scores and ranged from 0.46 for openness to experience to 0.76 for the positive valence factor. Our analysis thus suggested that both positive and negative item keying factors are heritable but acquiescence bias is not. Neither acquiescence bias nor item keying effect is found to be significantly influenced by the environment shared between the twins.
Scale Internal Consistency and Univariate Genetic Results for Final Model (M 2m+r)
Note: r = intraclass correlation; pos = positive valence factor; neg = negative valence factor; acq = acquiescence bias; MZ = monozygotic; DZ = dizygotic.
Values in parentheses represent the 95% confidence interval.
Finally, we examined whether controlling for the two response styles would substantially affect the heritability estimates of HEXACO personality factors. We compared estimates of the personality factors between the baseline model (M baseline; Table 3) and our final selected model (M 2m+r; Table 2). The estimates do not differ substantially. The extraction of method factors in M 2m+r does not substantially change the heritability estimates of personality traits.
Univariate Genetic Results for Baseline Model (M baseline)
Note: r = intraclass correlation.
Values in parentheses represent the 95% confidence interval.
Discussion
The current study advanced previous research in at least three significant ways. First, we were able to show both acquiescence bias and item keying response styles in our personality data. Previous researchers often examined and statistically controlled for only one of these response styles (e.g., Billiet & McClendon, Reference Billiet and McClendon2000; Marsh et al., Reference Marsh, Scalas and Nagengast2010; Motl & DiStefano, Reference Motl and DiStefano2002; Quilty et al., Reference Quilty, Oakman and Risko2006), and rarely attempted to model both biases simultaneously. An important advantage of using MTMM SEM techniques is that it allows us to model bias within the same framework and examine whether one type of bias can be fully explained by another bias. Our findings, however, suggested that both types of bias must be modeled to achieve the best fit to the data. In this way, our findings suggest future researchers should statistically control for both types of bias in their research, even though these response biases may not affect heritability estimates.
Second, an important advantage of modeling both response styles simultaneously is that it allows us to compare the magnitude between acquiescence bias and item keying, a comparison that is rarely considered. Although we found both acquiescence bias and item keying response bias within the dataset, their magnitude of influence on participants’ responses differed substantially. Item keying response styles explained 5.90% of the variance while acquiescence explained 2.11%. This result suggests that item keying affects variance in participants’ responses more than twice that of acquiescence bias in our sample. Although some researchers have suggested that response styles may not substantially affect the validity of a construct score (Schimmack et al., Reference Schimmack, Böckenholt and Reisenzein2002), the current research does not support this position. Variance explained by the two types of response bias (7.84%) is over one-fifth of the variance explained by personality constructs (33.18%). Consequently, response styles are substantive parts of construct measurement.
Third, most importantly, we found that the item keying effect in personality is heritable whereas acquiescence bias in personality is not. Past research by Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) only tested the heritability of item keying effect in relation to one construct (i.e., optimism). As we have explained, the reverse-keyed (negatively valenced) items of their optimism scale may actually represent the measurement of another construct (pessimism). However, with the personality constructs from HEXACO, ambiguity in the item keying effect is mitigated. Each personality dimension included in the current study is theoretically a unidimensional construct (Paunonen & Hong, in press; Wiggins, Reference Wiggins1973). In addition, we have included multiple measures in the study. For these reasons, the response styles detailed in the current study are unlikely to represent an opposite pole of any one particular personality construct, but rather a response bias on oppositely valenced personality items in general. Our methodology then overcomes a major limitation of Alessandri et al.'s (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010) study, which relies on only one construct in their investigation.
Perhaps due to the methodological improvement of the current study, our results differ substantially from Alessandri et al. (Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010), who found that only the positive valence factor in an optimism measure is heritable. When we examine positive and negative valence effects in univariate twin analyses, we find that both of these factors are heritable. In addition, the heritability was 24% for the positive valence factor and 33% for the negative valence factor, and both factors were not influenced by the environment that is shared by the twins. As compared with the heritability of personality traits in the current study (30–54%), our results thus suggest that the heritability of positive and negative valence factors is at the lower end of personality traits. This finding regarding the heritability of item keying is extremely important because it implies that participants’ responses to valence represent something that would be temporally stable and thus probably meaningful as it may be related with human biological system.
Contrary to the item keying response styles, our results revealed that acquiescence bias has little or no heritability. When we compared the correlation of acquiescence bias between MZ twins and DZ twins, we find a stronger relation for the former group. In spite of this, the univariate genetic analyses failed to find statistical significance in its heritability estimate. This result thus suggests the possibility that acquiescence bias is at least not as heritable as personality factors and item keying effects. The only significant influence on acquiescence bias was environmental factors that are not uniquely shared by the twins. This finding is interesting and warrants further investigation.
As with previous researchers (Alessandri et al., Reference Alessandri, Vecchione, Fagnani, Bentler, Barbaranelli, Medda and Caprara2010), we are still uncertain why item keying response styles are influenced primarily by heredity rather than shared environment (DiStefano & Motl, Reference DiStefano and Motl2006; Quilty et al., Reference Quilty, Oakman and Risko2006; Rauch et al., Reference Rauch, Moosbrugger and Schweizer2007). In contrast, past research on social and cultural psychology may already provide answers to the non-heritability of acquiescence bias. Cross-cultural psychologists have consistently showed that acquiescence is amenable to influences of external cultural factors. For example, there is overwhelming evidence that respondents from collectivistic cultures (e.g., East Asians) are more likely to acquiesce as compared to respondents from individualistic cultures (e.g., North Americans; Chen et al., Reference Chen, Shin-ying and Stevenson1995; Johnson et al., Reference Johnson, Kulesa, Cho and Shavitt2005; Smith, Reference Smith2004; van Hemert et al., Reference van Hemert, van de Vijver, Poortinga and Georgas2002). Collectivistic cultures encourage individuals to fit into an existing social structure and to conform to the overall goals or wishes of other members in the society. This experience affects their communication styles — they are more likely to acquiesce because doing so promotes relationship harmony and avoids conflict with others (Johnson et al., Reference Johnson, Kulesa, Cho and Shavitt2005; see also Kwan et al., Reference Kwan, Bond and Singelis1997; Peng & Nisbett, Reference Peng, Nisbett, Shield and Shale1996; Smith & Fischer, Reference Smith, Fischer, van de Vijver, van Hemert and Poortinga2008 for a discourse on this topic). If we look at our data closely, the results show the possibility that acquiescence is more amenable to shared environment between the twins than do other personality factors (c 2 = 0.09 vs. 0.00), although one may caution not to overinterpret this finding as its confidence interval for acquiescence overlaps with those for other personality factors. Given that our twin sample is mature in their age (90% of the twins are between the age of 34 and 77; mean age = 58.56 years), there are plenty of unshared, external environmental influences (i.e., the ‘E’ component in the model) on the twins once they live apart. The non-significance of the ‘C’ component for acquiescence may suggest that this bias is amenable to external influences even later in life, just as a person can still be adapted to cultures at least after adolescence (Oyserman & Lee, Reference Oyserman and Lee2008). The non-significance of heritability (i.e., the ‘A’ component) implies that biological factors are not central to the development of this response style.
Limitations and Future Directions
As with any other empirical studies, the current study has several limitations. First, our sample represents older twins from Britain and little is known regarding the generalizability of our findings to younger twins. Future research may try to replicate our results with younger twins or with people from other cultural backgrounds. Second, the sample is predominantly female so it would also be important to replicate the results with a more evenly sex-balanced sample. Third, although we included constructs from a commonly used personality model (HEXACO), we do not know whether item keying response styles from constructs outside personality are heritable. Therefore, confirmation of these results with other psychological constructs will be valuable. Finally, the current study investigates participants’ response styles to a 5-point Likert scale. Little is known whether the result will be generalizable to other response formats such as a 7-point Likert scale format or a true–false format. Although we do not expect the results will differ based on the scaling structure, replications of our findings in these other formats will be important.
Our study suggests researchers should shift their thinking regarding acquiescence bias and item keying response styles as representing something unworthy of empirical study. The fact that the heritability estimates of item keying response styles are at a similar level of some personality traits is itself intriguing and is worthy of further exploration. Acquiescence bias, although not as heritable as item keying response styles or personality factors, was found to be largely shaped by unshared environmental influences. Future research should continue to explore the nature of item keying response styles and acquiescence bias, because they are potentially important psychological phenomena that inform our understanding of human cognition and behaviors.
Acknowledgments
We would like to thank the twins for their voluntary contribution to this research project. We would also like to thank the staff of the Department of Twin Research for their help and support in undertaking this project. The Wellcome Trust provides core support for DTR. The study also receives support from the Department of Health via the National Institute for Health Research (NIHR) comprehensive Biomedical Research Centre award to Guy's and St Thomas’ NHS Foundation Trust in partnership with King's College London. Finally, we thank John Meyer for his valuable feedback on an earlier version of the manuscript.
