Microsoft Word ji job Pres Preprint docx

 
Knowledge hiding. We used Connelly et al.’s (2012) 4-item evasive knowledge hiding 
scale. Items were rated on a 7-point Likert type scale from 1 (Not at all) to 7 (To an extremely 
large extent) (
a = .88 T1, .87 T2, .90 T3).
Analytical Approach 
Confirmatory factor analyses (CFAs) and latent cross-lagged panel models were 
conducted using 
Mplus Version 8 (Muthén & Muthén, 1998-2017)
. To account for missing data, 
we used full information maximum likelihood (FIML) estimation, which has been shown to 
provide unbiased parameter estimates for longitudinal studies with missing time points (e.g., 
Enders & Bandalos, 2001). We used the following fit statistics as criteria: information criteria 
such as the Bayesian information criterion (BIC), chi-square difference tests and model fit 
indices including the comparative fit index (CFI), the root mean square error of approximation 
(RMSEA), and the standardized root mean squared residual (SRMR). A satisfactory fit was 
indicated by relatively lower BIC and conventional cut-offs such as 0.90 for CFI, 0.06 for 
RMSEA, and 0.08 for SRMR (Bentler, 1990). CFA was used to establish the validity and 
temporal invariance for the measurement model. Using a full longitudinal panel design, the latent 
cross-lagged model with paths for causal and reversed effects was conducted to test bi-
directional relationships between JI and workplace behaviors.
STUDY 1: RESULTS 
Table 2 displays the means, SDs, Cronbach’s alphas, and bivariate correlations.
-------------------------- 
Insert Table 2 about here 
-------------------------- 
CFA 
CFA models with varying numbers of factors were used to examine the discriminant 

---

JOB INSECURITY AND JOB PRESERVATION 13
validity of our measures. Consistent with Meier and Spector (2013) who used the same 
organizational counterproductive work behavior scale (Bennett & Robinson, 2000) in cross-
lagged panel models, we created four 3-item parcels as indicators for the construct of 
organizational counterproductive work behavior using the item-to-construct balance approach 
(Little, Cunningham, Shahar, & Widaman, 2002). Factors of all three times were correlated with 
each other, and residuals of indicators for the same construct were correlated over time (Meier & 
Spector, 2013). Consistent with previous research (e.g., Meier & Spector, 2013), the residual 
covariances were retained in the subsequent analyses. Factor loadings were freely estimated. 
This model (i.e., 15-factor model with JI, performance, counterproductive work behavior, 
evasive knowledge hiding, and self-presentation ingratiatory behavior at three times) was tested 
against three alternative models. As seen in Table 3, only the 15-factor model evidenced a 
satisfactory fit: CFI = .92, RMSEA = .051, SRMR = .055, and the lowest BIC among the three 
models. The 15-factor model fit statistically better than the three alternative models (p < .01).
-------------------------- 
Insert Table 3 about here 
-------------------------- 
To examine the effect of common method variance (CMV), we conducted the common 
method factor technique (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). In particular, we 
added the common method factor to the 15-factor CFA model. All items were allowed to load on 
an unmeasured common method factor, the path coefficients of the common method factor were 
constrained to be equal and the variance of the common method factor was fixed to be 1. The 
model with common method factor had larger chi-square and BIC, equivalent CFI, larger 
RMSEA and SRMR (common method factor model: χ
2
(1376) = 2719.49, BIC = 53854.00, CFI 
= .92, RMSEA = .052, SRMR = .063), and the factor loadings of items between the model with 

---

JOB INSECURITY AND JOB PRESERVATION 14
and without the common method factor did not change.
2