Search

A simulation of over one million samples from 5,184 population parameters was conducted to examine the performance of fit indices and rules of thumb in evaluating model fit. Results recommend a two-index strategy -- Standardized Root-Mean-Square Residual (SRMR) and C1 Gamma Hat -- and their cutoff values at various model parameter combinations. [ABSTRACT FROM AUTHOR]

Recent development in the area of measurement equivalence/invariance (ME/I) has promoted the prevalence of ME/I in organizational research. However, several questions concerning the practice of ME/I remain unclear or have not reached much consensus and cause confusion for organizational researchers. In this article, the authors discuss and provide answers to eight general questions concerning tests of ME/I. [ABSTRACT FROM AUTHOR]

Many statistical techniques were originally invented for the purpose of making comparisons between and among groups of entities. Not surprisingly, difficulties arise when the entities are human beings. Comparisons among groups of people are straightforward when the variables have objective operationalizations, such as age, income, and education level. They are less straightforward when the variables are attitudes, such as self-esteem and job satisfaction. Attitudes are usually operationalized as scale scores, and before such scores can be meaningfully compared, it must be shown that the scale items "mean the same thing" to members of all the groups being studied. This is the so-called measurement invariance problem. In recent years, structural equation modeling (SEM) has been used to test the measurement invariance of survey-type items across two groups; e.g., two national cultures. Unfortunately the currently recommended procedures are computationally intensive and not well suited for testing item invariance among multiple groups. In particular, they offer no straightforward ways of answering the following questions: (1) What subset of groups can be compared using all the items? (2) What subset of items can be used to compare all groups? This paper proposes a new technique, based on the confidence intervals of estimated factor loadings, which address these shortcomings. The procedure minimizes the number of SE models that must be estimated, and provides a systematic approach to analyzing the data produced. With the exception of the SEM runs, the entire procedure can be performed either manually or with the help of a spreadsheet program. [ABSTRACT FROM AUTHOR]

Confirmatory factor analysis is used to test for measurement artifacts in study of self-appraisal accuracy, including nonequivalent conceptualization of performance and dissimilar psychometric properties of measurement scales between self-ratings and supervisor ratings. Necessity for measurement equivalence to draw valid inferences of differences and similarities between self-rated and supervisor rated performance is highlighted. [ABSTRACT FROM AUTHOR]

Drawing on broaden-and-build theory and threat-rigidity hypothesis by Rhee (2007), we theorized and tested a multilevel model that examined the moderating effects of transformational leadership (TFL) on the team-level process linking group positive/negative affective tone (PGAT/NGAT) to team innovation via information elaboration. Data were collected from 299 team members and 65 leaders in Taiwan companies at two time points. The multilevel path analysis demonstrated support for a positive indirect effect from PGAT to team innovation via information elaboration and a negative indirect effect from NGAT to team innovation via information elaboration. In addition, the positive relationship between PGAT and information elaboration, and the indirect effect of PGAT on team innovation via information elaboration were stronger when TFL was high rather than low. However, TFL unexpectedly did not attenuate the negative effects of NGAT. NGAT was negatively related to information elaboration when TFL was high, whereas NGAT had no relationship with information elaboration when TFL was low. Theoretical and practical implications are discussed." [ABSTRACT FROM AUTHOR]

Despite the increasing popularity of studying higher-level constructs in organizational research in recent years, few empirical studies have clarified the conceptual, measurement, and statistical properties of their higher-level constructs. Based on the seminal work by Kozlowski and Klein (2000), we first summarize the differences in the conceptualization of the three basic types of higher-level constructs: constructs with global properties, shared properties, and configural properties. Then we differentiate the measurement properties for constructs with different conceptualizations. Finally, we demonstrate with simulation results that the multilevel latent covariate (MLC) approach is more appropriate for estimating higher-level constructs with shared properties while the multilevel manifest covariate (MMC) approach is more appropriate for estimating higher-level constructs with configural properties. We suggest future organizational researchers should first clearly define and specify the conceptual properties of the higher-level constructs, then use appropriate measurement that matches the conceptualization of the constructs, and finally adopt the proper statistical approach to aggregate individual observations to higher-level variables. [ABSTRACT FROM AUTHOR]

Since Campbell and Fiske (1959) defined convergent validity and discriminant validity, the tests for convergent validity and discriminant validity have evolved from checking the "high" and "low" correlation coefficients in the multitrait-multimethod context to specific rules of thumbs suggested by Fornell and Larcker (1981) in a multitrait-monomethod context. In this study, a simulation was conducted to first evaluate the effectiveness of (a) the Fornell-Larcker criterion for convergent validity, which requires the Average Variance Extracted (AVE) greater than 0.5 and (b) the Hair et al. criteria for convergent validity, which require the AVE greater than 0.5, standardized factor loading of all items not less than 0.5, and composite reliability not less than 0.7. We also evaluate the effectiveness of (c) the Forell-Larcker criterion for discriminant validity, which requires the AVE of both constructs greater than square correlation between the two constructs, (d) the Bagozzi et al. criterion which requires the correlation between two constructs significantly less than unity, and (e) the Kline criterion for discriminant validity, which requires the correlation between two constructs less than 0.85. Results show that all these criteria are not very effective in evaluating convergent validity and discriminant validity and most of them have ignored sampling errors such that the conclusion cannot be inferred to the population of interest. Finally, we recommend (a) concluding convergent validity if AVE is not significantly smaller than 0.5 and standardized factor loadings of all items are not significantly less than 0.5. and (b) concluding discriminant validity if correlation between two constructs is not significantly larger than 0.7. [ABSTRACT FROM AUTHOR]

Moderated multiple regression (MMR) is the most frequently employed analytical method for testing for interaction effects in non-experimental studies. Previous studies have identified many methodological artifacts, such as measurement error, range restriction, correlation among predictors, small sample size and small effect size, that can affect the power of MMR in determining the significance of interaction effects. However, examinations of the effects of these methodological artifacts on the confidence intervals (CIs) of the interaction effects estimated from MMR are rare. We thus conducted a simulation to show that MMR produces biased point estimates and CIs of interaction effects, where the bias is larger with a larger sample size, larger effect size and lower reliability. The coverage rate of the 95% CIs for a moderate interaction effect yielded by MMR with N = 500 and reliability at 0.8 is just slightly greater than 50%. A second simulation was conducted to demonstrate that the latent moderated structural equations (LMS) approach, which estimates latent interaction effects while controlling for measurement errors, may be a solution to the problems inherent in MMR. The results show that LMS produce both an accurate estimate of and accurate CIs for interaction effects under various conditions. [ABSTRACT FROM AUTHOR]

Although studies have demonstrated a positive relationship between work engagement and performance, research needs to more fully explore the network of linkages involving the antecedents and consequences of engagement, including important questions of mediation and moderation. This study proposes a comprehensive model of such linkages. Using latent moderated structural equation modelling on a sample of 320 permanent full-time front-line employees in an Indian organization, we find that having a more supportive supervisor and better personal resources (measured here as core self-evaluations) are positively associated with the work engagement of employees. More engaged employees are then more likely to report that they craft their jobs, a relationship that is strengthened when employees enjoy greater autonomy. Greater job crafting is associated with better task performance. Reinforcing the importance of employee control in their working environment, the mediating role of job crafting between engagement and task performance is enhanced by greater autonomy and whether greater work engagement leads to better task or contextual performance is influenced by the degree of autonomy that employees enjoy. [ABSTRACT FROM AUTHOR]

Using Goffman's (1959) original dramaturgical theory of self-presentation, we argue that the effectiveness of impression management (IM) usage can only be determined when ratings from both the actor and the audience are considered. To test our arguments, we employ congruence analysis (Cheung, 2009). Our approach differs from the majority of extant IM research that employs measures of IM only from the actor's perspective. By incorporating assessments from the actor and the audience, we bring research on IM back to its theoretical roots and offer a rationale for the varied and inconsistent findings reported in the literature. Using a sample of 170 employees and their supervisors, we examine and compare IM ratings of subordinates' ingratiation, self-promotion, and exemplification from both the employee and supervisor. Additionally, we investigate the ability of those ratings, individually and together, to predict both subordinate and supervisor ratings of key organizational outcomes. Implications for practice and suggestions for future research also are provided. [ABSTRACT FROM AUTHOR]

Moderated multiple regression (MMR) is the analytical method most frequently used for testing interaction effects in non-experimental studies. The power of MMR in testing the significance of interaction effects has been heavily criticized, especially because most observed variables in behavioral research are measured with errors. Yet, examination of the confidence intervals (CIs) of interaction effects estimated using MMR is rare. The latent moderated structural equations (LMS) approach used here, which estimates the latent interaction effect while controlling for measurement errors, may offer a solution to the problems of MMR. An approximation method of LMS, which corrects for reliability when estimating the latent interaction, is also introduced. Simulations were conducted to compare the CIs of interaction effects produced by the three approaches. The results show that MMR produced a bias point estimate and CI of the interaction effects. Increasing the sample size adversely affected the accuracy of CIs produced by MMR. In contrast, LMS produced both an accurate estimate and CI of the interaction effect under various conditions. [ABSTRACT FROM AUTHOR]

Thanks to the theoretical and analytical frameworks developed by Preacher, Rucker and Hayes (2007) and further advanced by Hayes (2013), studies of moderated mediation effects (conditional indirect effects) are booming in recent years. Yet, to date the most popular analytical method for testing moderated mediation is the regression approach (including MODMED and PROCESS), which is based on observed variables that assume no measurement errors. It is well-known that measurement errors attenuate the estimated regression coefficients, but the problem is more serious when both the predictor and moderator in a moderated mediation model are measured with errors. Hence, we extend the analytical frameworks developed by Preacher et al. (2007) and Hayes (2013) to a latent variable model approach. We developed two SPSS macros to circumvent some of the limitations of Mplus in generating the bias-corrected bootstrap confidence interval for the regression coefficient of the interaction term of two latent variables. Results in the numerical example show that the latent variable model approach produces estimated effects that are close to the true values despite that the standard errors and the confidence intervals are larger due to the incorporation of measurement errors in the estimation. On the other hand, the observed variable regression approach underestimates the effects and produces inaccurate confidence intervals for some estimates, including the estimated moderated mediation effect, where the confidence interval does not include the true value. It may also result in inflated Type I error which makes interpretation of results difficult. [ABSTRACT FROM AUTHOR]