Bifactor Model of Personality Predictors of Leadership Potential in Vietnam

1

Bifactor Model of Personality Predictors of Leadership Potential in Vietnam

Nhung T. Hendy, Towson University

NgaP. Pham, National Academy of Public Administration, Vietnam

Michael D. Biderman, University of Tennessee – Chattanooga

Presented at Symposium “Identifying Leadership Potential through Personality Assessment: Cross-Cultural Findings” Chaired by Ronald Page – Assessment Associates International

2015 Annual Conference of the Society for Industry & Organizational Psychology: Philadelphia, April 23 – April 25, 2015

Author’s Note: We would like to acknowledge Theodore A. Hendy for his assistance in data entry for this project

Abstract

In this study, a bifactor confirmatory factor analytic (CFA) model with orthogonal factors was shown to explain the Big Five personality factors of the Page Work Behavior Inventory (Page, 2001-2009) better than a CFA with correlated factors using a sample of 217 upper level undergraduate and MBA students in Hanoi, Vietnam. Specifically, after removing the general factor variance, the factor scores of the Big Five demonstrated substantial convergent validity with the raw composite scale scores. Emotional Stability remained a significant predictor of leadership potential controlling for extraversion, conscientiousness, and openness to experience.

Bifactor Model of Personality Predictors of Leadership Potential in Vietnam

Introduction

Assessing leadership potential and allocating resources to those individuals for further development to fill leadership positions within organizations have received increasing attention among organizational researchers. In this study, we define leadership potential as leadership emergence, which are individual differences or “factors associated with someone being perceived as leader-like” (Hogan, Curphy, and Hogan, 1994, p. 496). This definition is also similar to the definition proposed by Chan and Drasgow (2001) in which they introduced a construct called Motivation to Lead (MTL). Chan and Drasgow (2001) referred to leadership potential as a construct consisting of observed leadership behavior that would be engaged in the future once the individual is in a leadership position. Whereas our definition of leadership potential views it as the potential to emerge as a leader in a group, other definitions, e.g., Chan and Drasgow (2001) view leadership potential as leader effectiveness.

Meta-analyses on personality predictors of leadership (both emergence and leadership effectiveness per Lord, DeVader, and Alliger, 1986) have established that the Big Five factors of extroversion, conscientiousness, openness to experience, and low neuroticism (or high emotional stability) as the four traits having consistent and significant relationships to leadership emergence and leadership effectiveness. For example, Judge and colleagues (2002) in their most recent meta-analysis found that the multiple correlation of the above four personality traits with leadership (they combined two criteria of leadership emergence and effectiveness into one overall leadership criterion) to be .48 (Judge, Ilies, Bono, & Gerhart, 2002). They went on to state that the trait theory of leadership was well supported by the Big Five personality model. They also noted Emotional Stability became a non-significant predictor whencontrolling for the other three traits together in a multiple regression equation because it displayed the highest correlation with the other traits (Ones, Viswesvaran, & Reiss, 1996).

The purpose of this study is to apply a bifactor model approach to measuring personality traits to improve the validity of the Big Five in predicting leadership potential. A bifactor model is one in which a general factor orthogonal to and in addition to the substantive factors influences all personality items in a questionnaire. This general factor may also be called a common method factor (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003) although there is evidence that it may actually represent a particular item content (Biderman, McAbee, Chen, & Hendy, 2015). Evidence supporting the bifactor model in personality testing was reported by Biderman, Nguyen, and Sebren (2008) in which they found conscientiousness was less contaminated when the bifactor model was applied. After removing the contaminating variance by applying the bifactor model to the conscientiousness item level data, they found the validity of conscientiousness increased from (r = .09) to (r = .20) in predicting undergraduate academic performance (Biderman et al., 2008, Table 2, p. 892).

------

Insert Figure 1 about here

------

Further evidencesupporting the use of bifactor models when representing Big Five data includesBiderman, Nguyen, Cunningham, and Ghorbani (2011) showing thatpartialling out the general factorvariance dramatically changed the correlations of Big Five factors with outcome variables (e.g., positive and negative affectivity). It is our expectation that after removing the general factor variance, Emotional Stability will become a significant predictor of leadership potential when controlling for other traits of extraversion, conscientiousness, and openness to experience.Based on Biderman et al. (2011), we expect that estimating the general factor using the bifactor model will improve the fit of factor analytic models to the data of the Page Work Behavior Inventory – PWBI (Page, 2001-2009).

H1:EFA and CFA bifactor models of the PWBI will have a better fit compared to a correlated factors EFA and CFA models.

Judge et al. (2002) meta-analytically estimated that the high inter-correlation among four personality factors: extraversion, conscientiousness, neuroticism, and openness to experience caused Emotional Stability to be a non-significant predictor of leadership in a multiple regression equation. This suggests a general factor influencing all 4 personality factors.

H2: After removing the general factor variance, neuroticism will still remain a significant predictor of leadership potential compared with the three traits of conscientiousness, extraversion, and openness to experience.

Method

Participants

Participants were 219 upper-level undergraduate and MBA students from Hanoi, Vietnam. They were recruited to participate in this study in exchange for a small compensation (US$ 2.50 per person). The mean age of participants in this sample was 21.36 ranging from 20 to 41 years of age. The vast majority of the sample was female(81.3%).

Procedure.Participants were asked to complete a paper-and-pencil questionnaire (Page Work Behavior Inventory or PWBI). The PWBI was translated from English to Vietnamese and back to English. Where minor differences in translations occurred, consensus was reached by the 1st and 2nd authors who are native Vietnamese speakers and fluent in English.

Measures

Big FivePersonality Factors. The Page Work Behavior Inventory (PWBI) is a self-report personality assessment questionnaire designed to assess work styles and leadership potential. According to the Manual, the PWBI consists of 240 items making up 40 scales, of which 21 scales tap the Big Five personality traits (Page, 2001 – 2009). Four scales of sociability, leadership, influence, and energy make up Extraversion. Three scales of cooperation, concern for others, and diplomacy make up Agreeableness. Six scales of achievement orientation, Initiative, persistence, attention to detail, dependability, and rule following make up Conscientiousness, three scales of self-control, stress tolerance, self-confidence, and emotional awareness make up Emotional Stability; and four scales of adaptability, innovation, analytical thinking, and independence make up Openness to Experience. Each of the twenty-one scales making up the Big Five is measured by 10 items. Sample items include: “It is easy for me to meet people and make new friends” (Extroversion); “I make special efforts to be helpful and supportive of others” (Agreeableness); “I have an active imagination that creates many new ideas” (Openness to experience); “Once I reach a goal, I immediately set one that stretches me further” (Conscientiousness); “I don’t get emotional in tense situations” (Emotional Stability). The internal consistency estimates of the subscales that make up the Big Five personality factors are shown in Table 2. Compared to the U.S. samples, the reliabilities of the PWBI data in this Vietnamese sample were low. A discussion of the potential reasons for the low reliabilities is offered later in the paper.

Leadership potential. This variable was measured by sixty items measuring thirteen dimensions selected from the 18 subscales making up four of out the Big Five personality factors. Sample items include “I usually take charge of things when I am in a group”; “I am quite effective at bargaining and negotiating with difficult people”; “I often volunteer and start projects that others usually avoid”. The internal consistency estimate of this variable was .87.

Analyses.All statistical analyses were conducted using SPSS version 21.0 and MPlus version 7.3 (MuthénMuthén, 1998 - 2014). Both Exploratory Factor Analysis (EFA) using the Exploratory Structural Equations Modeling (ESEM) procedure in Mplus and Confirmatory Factor analysis (CFA) were used. The reason we conducted an EFAusing ESEM was to remove any doubt that the variance explained by the general factor could be better explained by cross-loadings. Typically, cross-loadings account for variance in a multidimensional construct, but analysis with an EFA can show that the general factor accounts for common variance beyond that accounted for by the cross-loadings (e.g., Morin, Arens, & Marsh, 2014).ESEM is a statistical technique that combines Exploratory Factor Analysis (EFA) and targeted rotation that allows the researcherto specify the general nature of the solution for the data.

Results

Table 1 shows the descriptive statistics and inter-correlations of variables included in the study. As shown in Table 1, all five factors of the Big Five – raw scale scores - were positively and moderately correlated with one another (mean correlation = .476). Further, all five variables were highly and positively correlated with leadership potential (mean correlation = .716). Table 2 shows the internal consistency estimates of all twenty-one subscales that make up the Big Five factors of personality.

------

Insert Tables 1 and 2 about here

------

As shown in Table 2, most subscales of the Big Five personality factors had reasonably good reliabilities (closer to or above .7) except for diplomacy (.56), leadership (.27), and emotional awareness (.45). We offer some possible reasons for the low reliabilities of the above subscales in the discussion section.

Hypothesis 1 states that EFA and CFA bifactor models of the PWBI data will have better fit compared to corresponding correlated factors models without the general factor. To test this hypothesis, a total of five models were tested to validate the factor structure of the PWBI data in this study. Model 1 is an ESEM Exploratory Factor Analysis with five factors allowed to be correlated with one another. The five factors are Extraversion, Agreeableness, Conscientiousness, Emotional Stability, and Openness to Experience. The number of indicators for these five factors are 4, 3, 4, 6, and 4 (total = 21) respectively. Model 2 is an ESEM Bifactor Exploratory Factor Analysis model. This model is similar to Model 1, but with an additional factor, called the General Factor influencing all twenty-one indicators making up the Big Five factors. Model 3 is an ESEM Bifactor model EFA with five factors constrained to be uncorrelated or orthogonal to each other. Although the fit of EFA models with orthogonal factors is identical to EFA models with the same number of oblique factors, we included both in order to examine the convergent and discriminant validity of the factors in the oblique factors model. Model 4 is a Confirmatory Factor Analytic (CFA) model with five factors allowed to be correlated with one another.Model 5 is a Bifactor CFA model with orthogonal factors.This model is similar to Model 4, except that it includes a sixth factor, called the General Factor, influencing all twenty-one indicators.We used a variety of fit indices to evaluate model fit including Comparative Fit Index (CFI); Tucker-Lewis Index (TLI); Root Mean Square Error of Approximation (RMSEA); and Standardized Root Mean Square Residual (SRMR). Whereas models having CFI and TLI values of .9 or above are considered a good fit to the data; models with RMSEA and SRMR values close to .06 and .08 respectively are considered having a good fit (HuBentler, 1999). Model comparison was conducted using the chi-square difference test(Yung, Thissen, & McLeod, 1999.)

Model 5 was retained as the best fit model to the PWBI data in this Vietnamese sample based on all fit indexes shown in Table 3 and considerations related to whether items loaded on the appropriate factors. Table 4 shows factor loadings and factor score correlations across estimated models. As shown in Table 4, the results of the ESEM EFA, while supporting the existence of a general factor (as shown in the significant loadings of all indicators on to the general factor per Table 4) were not wholly satisfactory (loadings on the Openness to Experience factor was not significant in Models 2 and 3; loadings on the Conscientiousness factor was not as expected in Model 4), therefore we decided to retain Model 5 as the best fit model to the PWBI data. Thus, Hypothesis 1 was fully supported.

------

Insert Tables 3, 4, 5, & 6 about here

------

Table 5 shows the inter-correlations of the Big Five factor scores after removing the general factor variance. As shown in Table 5, the correlations among the Big Five factors with the general factor variance removed were substantially lower and near zero in magnitude. A comparison of Table 1 and Table 5 reveals that the general factor clearly influenced all five of the Big Five factors such that when the raw composite scale scores were used (per Table 1), all five factors were substantially and positively correlated with one another (mean correlation = .476). However, after removing the general factor variance, the mean correlation of the Big Five factor inter-correlations reduced to .09. This was much closer to what the scale developer/author had in mind – an independent factor structure of the Big Five as assessed by the PWBI. It is important to note that the factor scores from the CFA with orthogonal factors are essentially uncorrelated while at the same time had a better fit than the CFA with correlated factor (we did not run a Bifactor CFA model with correlated factors due to the small sample size resulting in unstable parameter estimates). This means that these factor scores will not be subject to the problems of collinearitythat Judge et al. (2002) as well as Ones et al. (1996) reported.

Table 6 shows the convergent validity of the factor scores from Model 5, the Bifactor CFA with orthogonal factors model with the scales scores from PWBI data. As shown in the Table, the Big Five scores derived from the Bifactor CFA model with orthogonal factors were not much different from the raw scale scores. In fact, they were all very highly correlated (mean correlation = .69). Further, the general factor scores can be estimated from the mean of all scale scores (21) making up the Big Five factors in the PWBI. This was evidenced in the near perfect correlation between the general factor score and mean of scale scores (r = .945)

Hypothesis 2 states that after removing the general factor variance, Emotional Stability will still remain a significant predictor of leadership potential compared with the three factors of conscientiousness, extraversion, and openness to experience. To test this hypothesis, we computed factor scores of four Big Five factors (i.e., extraversion, conscientiousness, emotional stability, and openness to experience). Because of the unexpected factor loadingsonto the Conscientiousness and Openness to experience factors estimated in Models 3 and 4 using ESEM, we created factor scores from the CFA solution with orthogonal factors rather than the ESEM solution. All factor scores were without bifactor variance and were imported into SPSS for the multiple regression analysis. We regressed the leadership potential composite scale score onto extraversion, conscientiousness, emotional stability, and openness to experience concurrently (meaning they were entered into the regression equation at the same time). The results are shown in Table 7.

------

Insert Table 7 about here

------

As shown in Table 7, all four predictors were statistically significant, explaining 12.4% of variance in leadership potential. The multiple R was .375. Further, Emotional Stability remained a significant predictor (β = .195, p < .001) with the other three Big Five factors in the regression equation. Thus, Hypothesis 2 was fully supported.

Discussion

This study was the first to validate the factor structure of the PWBI using a Vietnamese sample. We found that the Big Five Factor structure emerged, albeit, with room for further improvement in reliability. For example, two subscales: leadership and emotional awareness had very low internal consistency estimates (.27 and .45 respectively). An examination of the items reveals potential reasons for the low reliabilities: evaluative content of the items; complexity of the items, and translation accuracy. The evaluative content of items in the leadership subscale includes “I don’t enjoy giving orders and insisting that people do certain things”. Another example of an item with evaluative content in the emotional awareness subscale is “When I’m emotional, I know the reasons for all my feelings”. According to Saucier (2009), evaluative content of items designed to measure personality should be avoided due to “concern… that highly evaluative terms… might produce factors reflecting patterns of response to items with unusual extremeness in desirability rather than any meaningful substantive, descriptive content” (Saucier, 2009, p. 1581). In fact, despite the conscious effort to avoid using evaluative items from personality assessment, empirical evidence shows that the Big Five factors are usually correlated with one another, indicating an overriding factor of evaluative content (Block, 1995).Biderman, et al (2015) found that on average the correlations among raw scale scores were inflated from 20% to 40% with positive affect criteria and deflated about 20% with negative affectcriteria using NEO-FFI and HEXACO data.