0 Up votes0 Down votes

83 views33 pagesAsif khan Niazi Muju Islamabad

Apr 22, 2013

© Attribution Non-Commercial (BY-NC)

DOC, PDF, TXT or read online from Scribd

Asif khan Niazi Muju Islamabad

Attribution Non-Commercial (BY-NC)

83 views

Asif khan Niazi Muju Islamabad

Attribution Non-Commercial (BY-NC)

- QTBR
- !!! Methodology Mediation Moderation Emsley
- med_moderation_analysis
- Mediating moderation
- Linear Models
- Transformational Leadership and Work Engagement
- Century National Banks
- Mediation
- 1-s2.0-S1877042813036136-main
- Simmons Dissertation
- MacKinnon-Introduction to Statistical Mediation Analysis
- Beyond Baron Kenny Hayes2009
- Beta Just Dial
- Two-Stage Stochastic and Deterministic Optimization
- The Relationship Between Religious Attitudes and Perceptions of Organizational Justice in Coaches
- Statistics Lecture 2
- Additive models
- Corporate social responsibility
- Week 4 Jeopardy
- Business Statistics 1

You are on page 1of 33

: Preparing data 2nd stage: Exploring data 3rd stage: Describing data 4-stages of Quantitative Analysis II 4th stage: Analyzing data

Lecture 10:

In our incoming classes, we will visit special application-cases like: 1. Mediation & Moderation analysis 2. Panel data analysis 3. Time series analysis Unit roots analysis Cointegaration analysis Error Correction Modeling

Professor at Mohammad Ali Jinnah University, Islamabad (can be reached at anwar@jinnah.edu.pk; chishti_anwar@yahoo.com)

Baron and Kenny (1986) versus Kenny (2012) Mediator (M) is an intervening or process variable, and mediation analysis is the process analysis, that helps understand the mechanism through which the factor (X-variable) affects the outcome (Y-variable). Mediation analysis aims at to find whether the mediator M partially or totally mediates X - Y relationship; and if partially, then how much? (Wikipedia, 2012; Kenny, 2012; MacKinnon, 2008). Reuben Baron and David Kenny are considered among the major early pioneers who laid down foundations for extensive research in the area of mediation analysis. Their classic research article entitled The Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strategic, and Statistical Considerations, published in the Journal of Personality

and Social Psychology (Baron and Kenny, 1986), is considered one of the most-read papers,

with 15000 citations (Kenny 2012). This paper, for the first time, asked for making differentiation between Moderator and Mediator variables (Baron and Kenny, 1986). Most importantly, this article set the procedure which not only has been followed for mediation analysis over the last 2 3 decades, but is still being followed in its original shape by a vast majority of academia and researchers all over the globe. The procedure set for mediation analysis in the Baron and Kennys (1986) classical research article has not been without criticism (MacKinnon and Fairchild, 2009; Hayes, 2009; Bullock, Green, & Ha, 2010; Zhao, Lynch and Chen, 2010), and thanks to those critics, that Kenny (2012) had to bring a number of modifications and improvements in his todays contemporary mediation analytic procedure. The purpose of this paper is to present a comparison of what Baron and Kenny (1986) had originally proposed, and what Kenny (2012) has now suggested after incorporating critics concerns. For this purpose, the two approaches of meditational analysis, classical and contemporary, have been practically applied on an organizational justice - trust in supervisor - employees job satisfaction case, wherein variable trust in supervisor is being taken to mediate (as M-variable)

between the various facets of organization justice (X-variable) and employees job satisfaction (Y-variable). Organizational justice - trust in supervisor - employees job satisfaction The theory of organizational justice is concerned with the employee perceptions of the fairness of work-related issues; this concept has evolved over the years, from two dimensions in 1970s to three dimensions in 1980s and finally to four dimensions in the 1990s. Today, justice scholars generally agree that organizational justice is comprised of four major dimensions, namely distributive justice, procedural justice, interactional justice and informational justice. Its first facet, distributive justice, refers to the perceived fairness of decision outcomes such as pay, recognition, promotions, performance appraisal, and rewards. Employees compare the ratio of their inputs (efforts) and outcomes (rewards) to that of a referent employee. Procedural justice refers to the perceived fairness of the decision-making processes and procedures. Interactional justice refers to the respect and propriety of the relationship between employees, and their supervisors and managers, and the assessment that relationships are disrespectful or improper leads to perceptions of unfair treatment. Informational justice refers to the truthfulness and justification of information provided to employees, and the assessment that information is inadequate or untrue leads to perceptions of unfair treatment (Bies & Moag, 1986; Greenberg, 1990; Greenberg, 1993; Colquitt, 2001; Colquitt & Shaw, 2005). The experts in the area have found various facets of organizational justice linked with key organizational outcomes, including job satisfaction, organizational citizenship behavior, commitment, favorable assessment of supervisors, and trust. Perceived unfair treatment, in contrast, has been shown to lead to counterwork behaviors such as sabotage, intention of quitting, and antisocial behavior (Ambrose, Seabright and Schminke, 2002; Greenberg, 1997; Greenberg and Lind, 2000; Henle, 2005). Organizational justice has become such an important topic in organizational research that there have been more than 500 research articles written and more than 20 books devoted to this topic uo to 2005 (Colquitt, Greenberg & Scott, 2005). The discussion in the preceding section on various facets of organizational justice and its outcomes naturally brings up a question to the forefront: whether or not these facets of organizational justice prevail in Pakistani organizations, and if they do, then whether or not the levels of their prevalence are sufficiently enough to determine employees job satisfaction. Since, in almost all facets of organizational justice, supervisors are supposed to play positive role, trust4

in-supervisor is therefore included to test its role as mediator between organizational justice and job satisfaction. Research questions While pursuing the following research questions, we will demonstrate what has been the Baron and Kennys (1986) procedure of analyzing mediation, and what improvements Kenny (2012) has brought in the procedure. 1. Do the four facets of organizational justice prevail to the extent to determine employees job satisfaction in Pakistani organizations? 2. Does the trust in supervisor mediate between the various facets of organizational justice and job satisfaction?

As already discussed, Baron and Kenny (1986) have originally contributed the basic model for testing of the mediational effect of some variable, like the one we have introduced in our above given research question 2. To clarify the estimation procedure of mediation, the researchers introduced the following path diagram (Figure 1). Figure 1 (Adapted from Baron and Kenny, 1986) Mediator (M) a (Step 2) Independent Variable (X) b (Step 3) Outcome variable (Y)

c (Step 4) c (Step 1)

Baron and Kennys (1986) mediation analysis requires taking four distinct and consecutive steps for establishing mediation; these steps are (Judd and Kenny, 1981; Baron and Kenny, 1986; Kenny, 2012): First, initially in step 1, it is required to show that some initial variable (X) is correlated with the outcome variable (Y); that means estimating and statistically testing path c for H0: c = 0, in the above figure, suggesting that there is an effect (c) that may be mediated. Second, step 2 should show that the initial variable (X) is correlated with the mediator (M); that means estimating and statistically testing path a for H0: a = 0, suggesting to treat the mediator as if it were an outcome variable.

5

Third, step 3 should show that the mediator (M) affects the outcome variable (Y), while initial variable (X) is used as control variable; that means estimating and testing path b for H0: b = 0; suggesting that M-variable may mediate. Fourth2, step 4 is required to establish that M-variable completely mediates the X-Y relationship; that means estimating and testing path c the coefficient of X-variable while controlling for variable M; path c' should be zero for complete mediation, otherwise not. The Baron and Kennys (1986) four steps narrated above necessitate that: a. The relationships between variables X, M and Y be established through regression analysis; b. Statistical significance of the paths (c, a, b & c) be established using proper procedure, that is, evaluating via H0: c = 0; a = 0; b = 0; c = 0. c. In case c 0 in step 1, a 0 in step 2, b 0 in step 3 and c = 0 in step 4, there will be complete mediation; otherwise, in case of c 0 in step 4, there will be partial or incomplete mediation. Applying Baron and Kennys (1986) model We now apply Baron and Kennys (1986) analytic framework to establish whether trust in supervisor (TS) mediates the relationship between four facets of organizational justice distributional justice (DJ), procedural justice (PJ), interactional justice (IJ) and informational justice (INJ) and employees job satisfaction (JS). Figure 2 represents the case, with various paths to be evaluated as per discussion made earlier. Figure 2 Organizational justice - trust in supervisor - job satisfaction M = trust in supervisor (TS) a b

c c

The effects in both Steps 3 and 4 (b and c) are estimated in the same equation.

6

Econometric model The above graphic presentation of variables X Y relationship, via a mediator M, requires estimation of an econometric model specified, as follows. Step 1: JS Step 2: TS JS = a0 + a1DJ + a2PJ + a3IJ + a4INJ + e2 = c0 + c1DJ + c2PJ + c3IJ + c4INJ + bTS + e3 (2) (3) Step 3 & 4: = c0 + c1DJ + c2PJ + c3IJ + c4INJ + e1 (1)

Data collection measures/constructs In order to estimate the relationship specified in Figure 2 and econometric models 1 3, the required data were collected from 276 employees relating to both public and private sector organizations situated in Rawalpindi-Islamabad area, using the measures on JS, TS, DJ, PJ, IJ and INJ as parts of a self-administered Likert-scale questionnaire, provided in Annex I. Reliability test The respondents responses on the respective elements of all six measures (JS, TS, DJ, PJ, IJ and INJ) were tested for reliability, and the following Cronbachs Alphas were estimated (Table 1). Table 1 Results of reliability test Construct Cronbachs Alpha Job Satisfaction (JS) 0.739 Trust in supervisor (TS) 0.692 Distributive Justice (DJ) 0.828 Procedural Justice (PJ) 0.890 Interactional Justice (IJ) 0.920 Informational Justice (INJ) 0.834 According to Uma Sekaran (2003), the closer the reliability coefficient Cronbachs Alpha gets to 1.0, the better is the reliability. In general, reliability less than 0.60 is considered to be poor, that in the 0.70 range, acceptable, and that over 0.80 are good. The reliability tests of our measures/constructs happened to be in the acceptable-to-good and very-good range. After having

7

reliability tests of all measures/constructs in good ranges, data on elements of constructs were averaged row-wise to generate data on respective variables Job Satisfaction (JS), Trust in Supervisor (TS), Distributive justice (DJ), Procedural justice (PJ), Interactive justice (IJ) and Informational justice (INJ). Regression analysis and results According to the two research questions set earlier for this research, the researchers need to test the following respective hypotheses: Hypothesis: H1: The four facets of organizational justice prevail in Pakistani organizations to the levels that seem sufficiently enough to determine employees job satisfaction in Pakistani organizations Hypothesis: H2: Trust-in-supervisor plays mediating role between the four facets of organizational justice and employees job satisfaction. As per hypothesis H2, if one is interested to test whether a variable is mediating or not, then, according to Baron and Kennys (1986) model, a 3-step regression needs to be run, as discussed earlier. Doing so yielded the following empirical results: Step 1: JS = = c0 + c1DJ + c2PJ + c3IJ + c4INJ + e1 2.155 + 0.092DJ 0.010PJ + 0.071IJ + 0.278INJ (0.0445) (-0.042) (0.0385) (0.0658) (2.0450) (-0.237) (1.8420) (4.2260) (0.0420) (0.814) (0.0670) (0.0000) F = 21.055 (p = 0.000) R2 = 0.237 R2adjusted = 0.226 (4)

(Figures in the 1st, 2nd and 3rd parentheses, respectively, are standard errors, t-ratios & p-values) The estimated model 4 is found statistically significant (F = 21.055, p < 0.01). With the exception of variable procedural justice (PJ), all other three components of organizational justice statistically significantly contribute towards job satisfaction. Informational justice (INJ) has the greatest contribution (c4 = 0.278, p = 0.000), followed by distributive justice (DJ; c1 = 0.092, p = 0.042) and interactional justice (IJ; c3 = 0.071., p = 0.067).

Step 2:

8

TS

= =

a0 + a1DJ + a2PJ + a3IJ + a4INJ + e2 1.770 + 0.037DJ 0.042PJ + 0.063IJ + 0.322INJ (0.4551) (0.0437) (0.0396) (0.0673) (0.8130) (-0.962) (1.5890) (4.7850) (0.4170) (0.3370) (0.1130) (0.0000) F = 24.270 (0.000) R2 = 0.264 R2adjusted = 0.253 (5)

Model 5 is found statistically significant (F = 24.270, p < 0.01). However, with the exception of informational justice (INJ) variable, all other three components of organizational justice have turned out to be statistically insignificant. INJ variable has the greatest contribution ( a4 = 0.332, p = 0.000), followed by others with statistically negligible contributions. Step 3 (& 4): JS = c0 + c1DJ + c2PJ + c3IJ + c4INJ + bTS + e3 = 1.891 + 0.086DJ 0.016PJ + 0.062IJ + 0.230INJ + 0.150TS (0.0444) (0.0413) (0.0386) (0.0679) (0.0588) (1.9370) (-0.387) (1.608) (3.3870) (2.5500) (0.0540) (0.6990) (0.1090) (0.0010) (0.0110) F = 18.487 (0.000) R2 = 0.255 R2adjusted = 0.241 (6)

Model 6 is found statistically significant (F = 18.487, p < 0.01). According to step 3, the contribution of TS variable is substantial and statistically significant (b = 0.150, p = 0.011). In accordance with step 4, with the inclusion of variable Trust in supervisor (TS)), the contributions of DJ, IJ and INJ variables have decreased from c1 = 0.092, c3 = 0.071 and c4 = 0.278 (Step 1) to c1 = 0.086, c3 = 0.062 and c4 = 0.230 (step 3), respectively; however, the latter c have not turned out to be statistically equal to zeros the condition for complete mediation. This fulfils the condition for Trust in supervisor (TS) of being a mediator, and since the contributions of DJ and INJ variables are still statistically significant, the TS variable is therefore partially mediating. As far as hypotheses H1 and H2 are concerned, both hypotheses are accepted; however, the former one is fully accepted while the latter one partially.

The aforementioned analysis was carried out in accordance with Baron and Kennys (1986) paper; however, Kenny (2012) makes a number of noticeable changes in the Baron and Kennys (1986) earlier procedure of mediation analysis; these changes are discussed, as follows: 1. According to Baron and Kenny (1986), A variable functions as a mediator when it meets the following conditions:

9

(a) variations in levels of the independent variable significantly account for variations in the presumed mediator (i.e., Path a); (b) variations in the mediator significantly account for variations in the dependent variable (i.e., Path b); and (c) when Paths a and b are controlled, a previously significant relation between the independent and dependent variables is no longer significant, with the strongest demonstration of mediation occurring when Path c is zero. 2. According to Kenny (2012): (a) We note that Baron and Kenny (1986) steps are at best a starting point in a mediational analysis. More contemporary analyses focus on the indirect effect; (b) Note that the steps are stated in terms of zero and nonzero coefficients, not in terms of statistical significance; (c) Most contemporary analysts believe that the essential steps in establishing mediation are Steps 2 and 3, and not Step 1 and 4. Estimation of direct, indirect and total effect Kenny (2012), whereas asks for not emphasizing on statistical significance of the estimated coefficients (c, a, b & c), he gives more importance to measuring of total effect of X on Y through Path c, and its decomposition in to direct effect of X on Y through Path c and indirect (meditational) effect through a measure ab (product of a & b); hence: Total effect = Direct effect + Indirect effect c = c + ab (7a) (7b)

where c, c, a and b have already been introduced in the earlier sections. However, the equality of equation 7 holds in certain conditions and does not hold in others. This equation exactly holds in: (a) multiple regression and structural equation modeling (SEM) without latent variables; (b) when same cases are used in all the analyses; and (c) when the same covariates are used in all the equations. While the two sides of the equation are only approximately equal for multilevel models, logistic analysis and structural equation modeling with latent variables included. For the latter models, it is probably inadvisable to compute c from Step 1, but rather c should be inferred to be c + ab, and not directly computed (Kenny (2012). Since, in majority of the cases, the three coefficients ( c, a & b) would suffice, and for that, the required steps are steps 2 and 3, and not steps 1 and 4. However, it seems necessary that the

10

product term ab, which measures indirect or mediational effect, be checked for its statistical significance, using measures like (i) checking of significance of coefficients a and b individually, (ii) Sobel test for mediation and/or (iii) bootstrapping. Applying Kennys (2012) procedure As discussed above, Kennys (2012) contemporary mediation analysis requires putting values of c and a and b coefficients in model 7, and solving it for total effect c, and then decomposing the total effect in to its direct (c) and indirect effects (ab), using the following formulas. Direct effect (%) = (c / c) x 100 Indirect effect (%) = (ab / c) x 100 (8a) (8b)

The values of c, a and b, required for substituting in models 7 and 8 (a b), are already available in estimated models 4 6; however, it should be noted that the contribution of variable PJ appears extremely insignificant (equal to zero), in all the three estimated models (4 6), relative to other three facets of organizational justice (DJ, IJ & INJ). Additionally, the coefficient of this variable (PJ) carries a negative sign, which makes it inconsistent candidate for mediation analysis; Kenny (2012) discusses a number of such inconsistent mediation cases in his paper. Table 2, after using formulas 7 and 8 (a b), provides direct and indirect effects, suggesting that variable DJ, IJ and INJ apparently exert 93.94 percent, 86.77 percent and 82.64 percent direct effect, respectively, while the indirect (meditational) effect of trust in supervisor relative to these variables estimates at 6.06 percent, 13.23 percent and 17.36 percent, respectively. Table 2 Total, direct and indirect effects Coefficients DJ IJ INJ A 0.037 0.063 0.322 B 0.15 0.15 0.15 c' 0.086 0.062 0.23 Ab 0.00555 0.00945 0.0483 c = (c' + ab) 0.09155 0.07145 0.2783 Direct effect (c/c) 0.93938 0.86774 0.82645 In % 93.94 86.77 82.64 Indirect effect (ab/c) 0.06062 0.13226 0.17355 In % 6.06 13.23 17.36

11

Whether the variable trust in supervisor (TS) significantly mediate towards the respective total effects of each of the three facets (DJ, IJ & INJ) of organizational justice, Kenny (2012) suggests to check the statistical significance of their respective indirect effects ab; there are a number of ways to do so, including the following ones. 1. Testing a and b separately 2. Sobel test 3. Bootstraping We carry out the first two tests here, as under. Testing a and b separately One way to test H0: ab = o, is to test a = 0 and b = 0; according to Kenny (2012), a number of other researchers, including Fritz and MacKinnon (2007) and Fritz, Taylor and MacKinnon (2012) strongly urge that researchers use this test in conjunction with other tests, such as Sobel test. Hypothesis H0: a = 0 has already been tested in Step 2; estimated model 5 indicates that p-value of a4 and a3, which relate to variables INJ and IJ, respectively, are 0.00 and 0.113, and that of a1, which relates to variable DJ, is statistically insignificant. Hypothesis H0: b = 0 has already been tested in Step 3; estimated model 6 indicates that p-value of b, relating to mediation-variable TS, is 0.011. On the basis of this approach, the indirect/meditational effect ( ab) of variable INJ appears to be strongly statistically significant, followed by variable IJ, which seems to be moderately significant, while variable DJ happens to have little meditational effect. Sobel test To check H0: ab = 0, Sobel test uses the following test statistic. Test: Zab = ab/sab (9a) which follows Z-distribution, that is, ab/sab will fall within 1.96 interval for an ab = 0; otherwise, it will fall outside of the stated interval. Where sab has to be computed, using formula: sab =(a2s2b+ b2s2a) Where sa and sb are the standard errors of a and b, respectively. Table 3 provides detailed computations done for Sobel test, using formulas given in 9 (a & b). Table 3 Computations for Sobel test

12

(9b)

Sobel test computati ons 2 a b2 sa sb s2a s2b a2s2b b2s2a sab Zab

DJ 0.001369 0.0225 0.045510 455 0.058823 529 0.002071 202 0.003460 208 4.73702E06 4.6602E05 0.007165 128 0.774584 912

IJ 0.003969 0.0225 0.039647 577 0.058823 529 0.001571 93 0.003460 208 1.37336E05 3.53684E05 0.007007 282 1.348597 125

INJ 0.103684 0.0225 0.067293 63 0.058823 53 0.004528 43 0.003460 21 0.000358 77 0.000101 89 0.021462 94 2.250390 44

Statistic Zab, computed for mediational effect (ab) of TS on variables DJ and IJ, falls within the 1.96 interval, and that of variable INJ outside of the interval, suggesting that the mediational effects in respect of the former two variables are statistically insignificant, and latter variable significant.

The purpose of this paper, as explained in Part I, has been to demonstrate and compare the applications of Barron and Kennys (1986) classical methodology of mediation analysis and the one Kenny (2012) has called contemporary mediation analysis. Part II and III of the paper, respectively, provide detailed applications of the two approaches, classical and contemporary, using the same case of organizational justice employees job satisfaction relationship via the meditational role of trust-in-supervisor. The two referred parts of the paper have explained the differences of the two approaches in detail; this part reproduces the differences in summarized form along with the explanation as to how and why the contemporary mediational analysis has certain edge over the classical one.

13

First, the classic approach required the estimation of the four paths ( c, a, b & c), through four steps (Steps 1, 2, 3 & 4) and three regression equations (like Equations 1 to 3 or Equations 4 to 6), and their testing for statistical significance. The contemporary mediation analysis has shortened the paths from four to three (a, b & c), requiring only two steps (Steps 2 and 3), and two regression equations, declaring the first equation as unnecessary. Second, the classic approach required that path b needs to become statistically significant and path c insignificant in step 3 & 4 for a complete mediation; contemporary approach asks for adding c with ab for determining total effect c, and then decomposing c in to direct effect (c / c) and indirect or mediational effect (ab / c). Third, contemporary approach requires that mediation effect ab needs to be tested for non-zero, using diagnostic methods, including (i) testing a and b for non-zeros, separately, (ii) running Sobel test and (iii) performing Bootstraping. Fourth, whereas classic approach aimed at solving for full or partial mediation in abstract form, the contemporary approach has the edge over the classic, in quantifying the mediation effect (ab). In case of our solved example, classic approach could only help to indicate that trust-insupervisor variable is partially mediating, while in case of contemporary approach, mediation effect of this variable for INJ facet of organizational justice was not only quantified (17.36%) and tested for its non-zero effect, but zero-effect of other two facets, DJ and IJ, were differentiated.

References

Baron, R. and Kenny, I (1986). The moderator-mediator variable distinction in social psychological research, Journal of Personality and Social Psychology.Vol.51, pp.1173-82 Bullock, J. G., Green, D. P, & Ha, S. E. (2010). Yes, But Whats the Mechanism? (Dont Expect an Easy Answer), Journal of Personality and Social Psychology, Vol. 98, No. 4, 550 Colquitt, J. A. (2001), On the dimensionality of organizational justice: a construct validation of a measure. Journal of Applied Psychology, 86 (3): 386-400. Colquitt, J.A., Conlon, D. O., Ng, K. Y., Wesson, M. J., & Porter, C. O. L. H.( 2001).Justice at the millennium: a meta-analytic review of 25 years of research. Journal of Applied Psychology, 86 (3): 425-445.

14

Colquitt et.al. (2001), Justice at the millennium: A meta-analytic review of 25 years of organizational justice research. Journal of Applied Psychology, Vol 86(3), Jun 2001, 425445. Colquitt, J.A. and Shaw,J.C. (2005), How should organizational justice be measured? In J. Green and J.A. Colquitt (Eds.), Handbook Of organizational justice (pp113-152). Mahwah: NJ: Lawrence Erlbaum Associate. Colquitte, J. A., Greenberg, J., and Scott, B.A. (2005), organizational justice: where dp we stand? In J. Greenberg and J.A Colquitt (Eds.) handbook of organizational justice (pp 589619). Mahwah: NJ: Lawrence Erlbaum Associate. Greenberg, J. (1986), Determinants of perceived fairness of performance evaluations. Journal of Applied Psychology, 71: 340-342. Greenberg, J. (1990b), Organizational justice: yesterday, today and tomorrow. Journal of Management, 16: 399-432. Greenberg, J. (1993b), The social side of fairness: interpersonal and informational classes of organizational justice. In R. Grjppanzino (Ed.), Justice in the Workplace: Approaching Fairness in Human Resource Management: 79-103. Hillsdale, NJ: Erlbaum. Greenberg, J. (1994), Using socially fair treatment to promote acceptance of a work site smoking ban. Journal of Applied Psychology, 79: 288-297. Greenberg, J., & Folger, R. (1983), Procedural justice, participation and the fair process effect in groups and organizations. In P. B. Paulus (Ed.), Basic Group Processes: 235-256. New York: Springer-Verlag. Greenberg, J., & Lind, E.A. (2000), The pursuit of organizational justice from conceptualization to implication to application. In C.L. Cooper And E.A. Locke (Eds.), I/O Psychology: what we know about theory and practice (PP 72-105). Hayes, A. F. (2009). Beyond Baron and Kenny: Statistical Mediation Analysis in the New Millennium. Communication Monographs, Volume 76 Issue 4, December 2009, p. 408420. Kenny, D. A. (2012). Mediation; available on: davidakenny.net, April 2012 Lee, H. (2000), An empirical study of organizational justice as a mediator of the relationship among leader-member exchange and job satisfaction, organizational commitment and turnover intentions in the lodging industries.

15

MacKinnon, D.P. and Fairchild, A.J. (2009). Current Directions in Mediation Analysis. Curr Dir Psychol Sci. 2009 February 1; 18(1): 1620. doi:10.1111/j.1467-8721.2009.01598.x. Niehoff, B. P., and Moorman, R. H., (1993), Justice as a mediator of the relationship between methods of monitoring and organizational citizenship behavior . Academy of management journal, 36(3), 527-556. Podsakoff, P.M., MacKenzie S.B., Moorman, R.H., Fetter, R., (1990), Transformational leader behaviors and their effects on followers trust in leader, satisfaction, and organizational citizenship behaviors, Leadership Quarterly, 1(2), p. 107-142. Raymond Loi., James, M.D., and Yang,J. (2009), Four Factor Justice and daily job satisfaction: A multilevel Investigation. Journal Applied Psychology, 94(3): 770-781. Saunders, M., Lewis, P. and Thornhill, A. (2011), Research Methods for Business Students. 5th Ed. Skkaran,uma (2003), Research Methods for Business: A skill building Approach. 4th Ed. Zhao, X., Lynch, J. G., Chen, Q. (2010). Reconsidering Baron and Kenny: Myths and Truths about Mediation Analysis. Journal of Consumer Research, 37, 197-206.

16

Section 1 Your Organization: Your gender: 1. Male Your age (in years like 40 years) Your education (actual total years of schooling ) Your area of specialization: Your job title in this organization: Working years in this organization: 2. Female

Section 2 Trust Scale in Supervisor (Podsakoff et al., (1990) Strongly disagree 1 Disagree = 2 Not disagree/neither agreed = 3 Agreed = 4 Strongly agreed = 5 (Please do not generalize, but tell about what is actually happening in your organization only) 1 2 3 4 1 I feel quite confident that my supervisor will always try to treat me fairly 2 My supervisor would never try to gain an advantage by deceiving workers 3 I have complete faith in the integrity of my supervisor 4 I feel a strong loyalty to my supervisor 5 I would support my supervisor in almost any emergency 6 I have a divided sense of loyalty toward my supervisor (R) Job satisfaction (Agho et al. 1993; Aryee, Fields & Luk, 1999) I am often bored with my job (R) I am fairly well satisfied with my present job I am satisfied with my job for the time being Most of the day, I am enthusiastic about my job I like my job better than the average worker does I find real enjoyment in my work Organizational Justice (Niehoff and Moorman, 1993) Strongly disagreed = 1 Slightly disagree = 2 Disagree = 3 Slightly more agreed = 6 Neutral (Not disagree/neither agreed) = 4 Strongly agreed = 7 Agreed = 5 5

1 2 3 4 5 6

17

Distributive justice items 1 1 2 3 4 5 1 2 3 4 5 6 My work schedule is fair I think that my level of pay is fair I consider my workload to be quite fair Overall, the rewards I receive here are quite fair I feel that my job responsibilities are fair Formal procedural justice items Job decisions are made by my supervisor in an unbiased manner My supervisor makes sure that all employee concerns are heard before job decisions are made To make formal job decisions, supervisor collects accurate & complete information My supervisor clarifies decisions and provides additional information when requested by employees All job decisions are applied consistently across all affected employees Employees are allowed to challenge or appeal job decisions made by the supervisor 1 2 3 4 5 6 7 8 9 Interactive justice items When decisions are made about my job, the supervisor treats me with kindness and consideration When decisions are made about my job, the supervisor treats me with respect & dignity When decisions are made about my job, supervisor is sensitive to my own needs When decisions are made about my job, the supervisor deals with me in truthful manner When decisions are made about my job, the supervisor shows concern for my rights as an employee Concerning decisions about my job, the supervisor discusses the implications of the decisions with me My supervisor offers adequate justification for decisions made about my job When decisions are made about my job, the supervisor offers explanations that make sense to me My supervisor explains very clearly any decision made about my job 2 3 4 5 6 7

Informational justice items Strongly disagree 1 1 2 3 4 5 Disagree = 2 Not disagree/neither agreed = 3 Agreed = 4 Strongly agreed = 5 1 2 3 4 5

Your supervisor has been open in his/her communications with you Your supervisor has explained the procedures thoroughly Your supervisor explanations regarding the procedures are reasonable Your supervisor has communicated details in a timely manner Your supervisor has seemed to tailor (his/her) communications to

18

19

Moderator: definition

Kenny (August 2011) states We begin with a linear causal relationship in which the variable X is presumed to cause the variable Y. A moderator variable M is a variable that alters the strength of the causal relationship. Although classically, moderation implies a weakening of a causal effect, a moderator can amplify or even reverse that effect. Complete moderation would occur in the case in which the causal effect of X on Y would go to zero when M took on a particular valueA key part of moderation is the measurement of X to Y causal relationship for different values of M. We refer to the effect of X on Y for a given value of M as the simple effect X on YDeciding which variable is the moderator depends in large part on the researcher's interest. For the earlier example in which gender moderates the effect of psychotherapy, if one was a gender researcher, one might say that psychotherapy moderates the effect of gender.

An example

If a researcher thinks that variable X determines Y, he/she specifies the model, as follows. Y = a0 + a1X + e results.

Model Summary Adjusted R Model 1 R .248a R Square .062 Square .053 Std. Error of the Estimate 12.75574 Durbin-Watson 1.855

(12.1)

Using data from Morse-II data-file, the researcher runs regression (12.1), and gets the following

ANOVA Model 1 Regression Residual Total Sum of Squares 1143.460 17409.846 18553.306 Df 1 107 108 Mean Square 1143.460 162.709 F 7.028 Sig. .009a

20

Coefficients Standardized Unstandardized Coefficients Model 1 (Constant) X B 15.552 6.558 Std. Error 1.222 2.474 .248 Coefficients Beta t 12.729 2.651 Sig. .000 .009

The researcher finds reasonably good results, but he/she then assumes that there may be another variable M, which may moderate the relationship between X and Y; so he changes the specification of his model, as follows. Y = i + aX + bM + cXM + e (12.2)

That is, for incorporating moderator M, he adds M, as well as, its interactional form (XM = X x M) in the model; lets check whether his idea of inclusion of moderator works. Use data from Morse II data-file, and run equation (11.2) regression. Please note coefficient c must be statistically significant for the moderator to work. Results are:

Model Summary Adjusted R Model 1 R .473a R Square .224 Square .202 Std. Error of the Estimate 11.71090 Durbin-Watson 1.922

ANOVA Model 1 Regression Residual Total Sum of Squares 4153.052 14400.254 18553.306 Df 3 105 108 Mean Square 1384.351 137.145 F 10.094 Sig. .000a

Coefficients Standardized Unstandardized Coefficients Model 1 (Constant) X B 15.366 3.924 Std. Error 1.154 2.340 .149 Coefficients Beta t 13.319 1.677 Sig. .000 .097

21

Coefficients M XM 1.374 .417 .304 .604 .403 .060 4.522 .690 .000 .492

The results have improve in terms of R 2 and F-statistic; the new variable M is significant, but its interaction term with X (that is, XM) is not significant, as per the requirement of the effect of the moderating variable. However, lets think how we would have interpreted the results if the interaction term was found significant. Reproducing the results in equation form: Y = 15.366 + 3.924X + 1.374M + 0.417XM M: Y/X = 3.924 + 0.417M the effect of X and M on Y. Estimating Descriptive Statistics:

Descriptive Statistics N Y X M 109 109 109 Minimum .00 -.42 -3.46 Maximum 30.00 .58 16.54 Mean 15.5517 .0000 .0000 Std. Deviation 13.10686 .49616 3.84339

(12.3)

Taking derivative of Y with respect to X (to check the effect of X on Y, in presence of moderator (12.4)

Putting values of M (mean values, as well as, other values we are interested in), we can arrive at

We may use one-standard deviation plus-minus range (or -1 0.00 +1 range) of M-variable to check the level effect of this variable on X Y relationship. The respective range (of M, using one SD) is estimated, as follows. For M variable, range is: = (-3.8434 0.000 3.8434) Evaluating the effect of X, using (11.4): Y/X = 3.924 + 0.417(M = -3.8434) = 2.3213 Y/X = 3.924 + 0.417(M = 0.000) = 3.924 Y/X = 3.924 + 0.417(M = 3.8434) = 5.5267 The effect of X on Y enhances when M increases. (12.6a) (12.6b) (12.6c) (12.5)

22

In case the respective range M is set at -1 0.00 +1,the results would be: Y/X = 3.924 + 0.417(M = -1) = 3.507 Y/X = 3.924 + 0.417(M = 0.000) = 3.924 Y/X = 3.924 + 0.417(M = 3.8434) = 4.341 The effect of X on Y enhances when M increases. (12.a) (12.7b) (12.7c)

Nonlinear moderation refers to effect of X changing as function of M, but that change is nonlinear. The typical way to estimate nonlinear moderation would be to estimate the following equation: Y = d + a1X + b1M + b2M2 + c1XM + c2XM2 (12.8) Nonlinear moderation can be tested by determining if c2 is different from zero. Note that M2 effects can only be estimated if M takes on at least 3 values) The effect of X in Equation (12.8) is: Y/X = a1 + c1M + c2M2 Where 11.9 would be interpreted in one of the following ways, namely: If c1 were and c2 were positive, then the effect of X on Y would be increasing as M increases, and this increase is increasing as M increases, accelerating (increasing with an increasing rate). If c1 were positive and c2 negative, then the effect of X on Y would be increasing as M increases, but this increase is declining as M increases, de-accelerating (increasing with a decreasing rate. If c1 were negative and c2 positive, then the effect of X on Y would be decreasing as M increases, but this decrease is declining as M increases, de-accelerating (decreasing with decreasing rate). If c1 were negative and c2 negative, then the effect of X on Y would be decreasing as M increases, but this decrease is increasing as M increases, accelerating (decrease with increasing rate). (12.9)

23

Other cases

There are four types of X and M variables, namely:

1. Both X and M are categorical variables 2. Both X and M variables are continuous. 3. X is continuous and M categorical. 4. X is categorical and M continuous. {Most common {cases

The (below given) selected readings from David A. Kenny (August 2011) would help us to understand as to how each of the case would be estimated and interpreted.

24

David A. Kenny (August 8, 2011) Basic Definitions We begin with a linear causal relationship in which the variable X is presumed to cause the variable Y. A moderator variable M is a variable that alters the strength of the causal relationship. So for instance, psychotherapy may reduce depression more for men than for women, and so we would say that gender ( M) moderates the causal effect of psychotherapy ( X) on depression (Y). Most moderator analysis measure the causal relationship between X and Y by using a regression coefficient. Although classically, moderation implies a weakening of a causal effect, a moderator can amplify or even reverse that effect . Complete moderation would occur in the case in which the causal effect of X on Y would go to zero when M took on a particular value. The reader might consult papers by Kraemer and colleagues (2001; 2002) for a related but somewhat different approach to defining and testing of moderators. Frazier, Tix, and Barron (2004) provide a very good introduction to the topic of moderation and Marsh, Hau, Wen, Nagengast, and Morin. (2011) for a more detailed discussion of the topic. A moderation analysis is an exercise of external validity in that the question is how universal is the causal effect. A key part of moderation is the measurement of X to Y causal relationship for different values of M. We refer to the effect of X on Y for a given value of M as the simple effect X on Y. Deciding which variable is the moderator depends in large part on the researcher's interest. For the earlier example in which gender moderates the effect of psychotherapy, if one was a gender researcher, one might say that psychotherapy moderates the effect of gender. Timing of Measurement Ideally the moderator should be measured prior to variable X being measured. So if X is manipulated, then M should be measured prior to X being manipulated. Of course, if M is a variable that does not change (e.g., race), the timing of its measurement is less problematic. It is possible, but quite complicated, but M can be both a mediator and a moderator (see Kraemer et al. (2001) for a different point of view. Moderator and Causal Variable Relationship If X is a manipulated variable, in principal, there should be no relationship between X and M. If X is not randomized, it might be correlated with M. Unlike mediation, there is no need for X and

http://davidakenny.net/cm/moderation.htm

25

M to be correlated and that correlation has no special interpretation. However, if X and M are too highly correlated, there can be estimation problems. Measurement of Moderation Generally, moderator effects are indicated by the interaction of X and M in explaining Y. The following multiple regression equation is estimated: Y = i + aX + bM + cXM Equation (1) The interaction of X and M or coefficient c measures the moderation effect. Note that path a measures the simple effect of X, sometimes called the main effect of X, when M equals zero. As will be seen, the test of moderation is not always operationalized by the product term XM. Given Equation 1, the effect of X on Y is a + cM. Thus, the effect of X on Y depends on the value of M. It is noted that the effect of X on Y equals zero when M equals a/c, which may or may not be a plausible value of M. Alternative Interpretations of Moderator Effects Finding that c is statistically significant does not prove moderator effects. One major worry is non-additivity. Consider the case in which the relationship between X and Y is nonlinear. For instance, X is income and Y is work motivation. Imagine that the relationship between the two is nonlinear such that if X is small the relationship is larger than when X is large. If age were tested as a moderator, the income-motivation relationship, then because younger workers make less money, we would find the moderator effect, that the income-motivation relationship is stronger for younger than for older workers. Another worry is the actual moderator may not be the moderator but some other variable with which the moderator correlates. For instance, if we find that gender is a moderator, the real moderator might be height, masculinity-femininity, expectations of others, or income. Unless the moderator is a manipulated variable, we do not know whether it is a true moderator or just a proxy moderator. Level of Measurement of the Variables It is presumed here and throughout that the outcome variable is measured at the interval level of measurement. Should the outcome be a dichotomy, logistic regression would need to be used (Hayes & Matthes, 2009). The remainder of the page is organized around the levels of measurement of the moderator and the causal variable. The causal variable, X, can either be categorical (typically a dichotomy) or a continuous variable. So for instance, X might be psychotherapy versus no psychotherapy (a dichotomy) or it might be the amount of psychotherapy (none, one month, two months, or six months; a continuous variable). Much in the same way, the moderator or M can be either categorical (e.g., gender) or continuous (e.g., age). Readers are encouraged to read the next two sections, even if they are more interested in one of the other cases, as many key concepts in mediation are discussed there. Categorical Moderator and Causal Variables When both X and M are dichotomous (i.e., each have two levels), we have a 2 x 2 design. So for instance, psychotherapy (therapy versus no therapy), might be more effective for women than for men. We denote the four cells as X1M1, X1M2, X2M1,and X2M2. To estimate the above regression

26

equation, we need to dummy code the moderator and the causal variable. So for instance, if we use codes of zero and one, then we have the following interpretations of the coefficients in the above multiple regression equation (Equation 1: Y = i + aX + bM + cXM). a the effect of X when M is zero (the simple effect of X when M is zero) b the effect of M when X is zero c how much the effect of X changes as M goes from 0 to 1 The focus is on c because it captures the moderator effect. If c is positive, then it indicates that the effect of X on Y increases as M goes from 0 to 1. If c is negative, then it indicates that the effect of X on Y decreases as M goes from 0 to 1. Obviously the interpretation of moderator depends very much on how X and M are coded. If effect coding (one value of X and M is 1 and the other value is 1) is used, the interpretation of the coefficients is as follows: a the effect of X averaged across M (i.e., when M = 0) b the effect of M averaged across X (i.e., when X = 0) c half of how much the effect of X changes as M goes from -1 to 1 Which particular coding method that is used is largely a matter of personal preference. The important thing is to know what coding system is used and interpret coefficients accordingly. Although coding affects the coefficients, it does not affect the inferential statistic for the test of the interaction (but it does affect the tests of main effects), the multiple correlation, the predicted values, and the residuals. It is generally inadvisable to trim out of the multiple regression equation the main effects if the interaction is present in the equation. Regardless which coding system is used, there are four means because the design is 2 x 2. If effect coding were used, the means would equal (where i is the intercept in the regression equation): Cell Coding Predicted Mean X1M1 X = -1; M = -1 i a b + c X2M1 X = 1; M = -1 i + a b c X1M2 X = -1; M = 1 i a + b c X2M2 X = 1; M = 1 i + a + b + c There might be an interest in the effect of the causal variable or X for each of the levels of the moderator or the simple effects of X. To estimate the simple effects, a different regression equation is run and in each we recode the moderator so that a given level is set to zero (Aiken & West, 1991). If we want to test the effect of X when the M = 1, the equation is run but M is not used but M = M 1. Coefficient b is now the simple effect of X on Y when M is 1, because when M = 1, M is zero. If X or M have more than two levels, then multiple dummy variables are needed (the number of levels less one), and moderation is tested by a set of product variables. If there are covariates (variables that cause Y and measured prior to Y), they can be entered into the equation. If the covariates are themselves considered to be moderators, then they would be

27

allowed to interact with X. Note that predicted values for the four cells would no longer exactly equal the mean for the cell and so they should be referred to as least squares means. Effect Size Measurement of Moderator Effects and Power Analysis One can use traditional measures of effect size in measuring moderator effects f2 (see below). However, what seems preferable is to use a d change where d is Cohen's d mean difference divided by pooled standard deviation. That is, we measure the d for each of the two levels of M and compare them. In computing the two ds, we should use the same standard deviation. For instance, we might state: The effect of psychotherapy on depression yields a d of 0.4 for men and a d of 0.7 for women or 0.3. Because both M and X are dichotomies, this d change measure is itself a d. Because the design is 2 X 2, the estimate of the moderator effects can be viewed as a difference between two means (X1M1 and X2M2 vs. X2M1 and X1M2). Using these two means a d can be computed and a power analysis can be undertaken. Categorical Moderator and Continuous Causal Variable An example of this case, M is race, X is a personnel test, and Y is some job performance score. Generally, it is assumed that the effect of X on Y is linear. It is also assumed (but it can be tested, see below) that the moderation is linear. That is, as M varies, the linear effect of X on Y might vary. Thus, the linear relationship increases or decreases as M increases. It is almost always preferable to measure the linear effect by using a regression coefficient and not a correlation coefficient. More Complex Specification Nonlinear moderation refers to effect of X changing as function of M, but that change is nonlinear. The typical way to estimate nonlinear moderation would be to estimate the following equation: Y = d + a1X + b1M + b2M2 + c1XM + c2XM2 Equation (2)

Nonlinear moderation can be tested by determining if c2 is different from zero. (Note that M2 effects can only be estimated if M takes on at least 3 values.) The effect of X in Equation 2 is a1 + (c1 + c2M)M which would be interpreted as follows: If c1 were positive and c2 positive, then the effect of X on Y would be increasing as M increases, and this increase is increasing as M increases, accelerating. If c1 were positive and c2 negative, then the effect of X on Y would be increasing as M increases, but this increase is declining as M increases, de-accelerating. If c1 were negative and c2 positive, then the effect of X on Y would be decreasing as M increases, but this decrease is declining as M increases, de-accelerating. If c1 were negative and c2 negative, then the effect of X on Y would be decreasing as M increases, but this decrease is increasing as M increases, accelerating.

28

Baron and Kenny (1986, page 1175) discuss alternative specifications of moderation. For instance, the moderator might act as a threshold variable and there would be no effect of the causal variable when the moderator is low, but at a certain value of the moderator the effect emerges. In this case, the moderator is no longer continuous, but rather it is dichotomized at the point of the threshold. The difficulty is that threshold point must be known a priori and cannot be obtained by a simple median split. [The value of M at which the effect of X on Y changes might be empirically determined by adapting an approach described by Hamaker, Grasman, and Kamphuis (2010)]. Effect Size and Power The most common measure of effect size in tests of moderation is f2 (Aiken & West, 2001) which equals the unique variance explained by the interaction term divided by sum of the error and interaction variances. When X and M are dichotomies, f2 equals the d2/4 where d is the d difference measure described above. Cohen (1988) has suggested that f2 effect sizes of 0.02, 0.15, and 0.35 are termed small, medium, and large, respectively. However, Aguinis, Beaty, Boik, and Pierce (2005) has shown that the average effect size in tests of moderation is only 0.009. Perhaps a more realistic standard for effect sizes might be 0.005, 0.01, and 0.025 for small, medium, and large, respectively. We note that even these values are "optimistic" given the Aguinis et al. (2005) review. If f2 is known, one can conduct a power analysis using a power analysis program. For instance, if f2 is assumed to be 0.025, one needs a sample size of 316 to have 80 percent power. Power for tests of moderation is very low when one or both of the variables are continuous (McClelland & Judd, 1993). Likely, the much greater interest in mediation over moderation is due to the low power in tests of moderation. Simple Effects There are three methods to determine simple effects. The first method is to estimate the simple effects using the regression equation. Using Equation 1 (Y = i + aX + bM + cXM), we solve for a + cM. For instance, if the moderation regression equation were 5 + 2X + 3M + 1XM and we wanted to estimate the effect of X when M is 2, that effect would be 2 + (1)(2) or 4. The second method is to re-estimate separate regression equation but transform M by subtracting 2 or M' = M 2. For this new equation, the effect of X refers to the case in which M is 2. This second method should result in the same answer as the first. The third method requires that M take on a few values. Separate regression equations would be estimated for each value of M. This method does not assume homogeneity of error variances and so it would likely produce estimates different from the previous two. Continuous Moderator and Categorical Causal Variable An example is that the socio-economic status moderates the effect of some intervention. One key issue is to center the variable of socio-economic status; i.e., make sure that zero is a meaningful value for the moderator. We may want to determine the effect of X for various levels of the moderator, M, i.e., simple effects. In principal, the values of M would be chosen using some sort conceptual rationale. For

29

instance, if IQ were the moderator, we might use 140 (genius level) and 100 (average level) to compute the effects of X on Y. More commonly, the values are one standard deviation above the mean of M and one standard deviation below the mean of M. To obtain these estimates we use either the first or second method described above. Continuous Moderator and Causal Variable One key question is the assumption of how the moderator changes the causal relationship between X and Y. Normally, the assumption is made that the change is linear: As M goes up or down by a fixed amount, the effect of X on Y changes by a constant amount. Alternatively, M may have a different type of effect: Threshold The effect of X on Y changes when M is greater than a certain value; Discrepancy When X and M are measured using the same units, the absolute difference between X and M is what matters (see also Edwards, 1995). The key point is that moderation is not always best captured by a product term. If a product term is used, one must assume that both X and M are measured without error, an often dubious assumption. Latent variables are discussed below. Centering of both X and M is necessary if neither have zero as a meaningful value. To interpret the results and determine simple effects, the effect of X at various levels of M would be measured. Ideally, the levels of M would be theoretically motivated. If not possible, one might use M at the mean and at plus and minus one standard deviation from the mean. Power of tests of moderation with two continuous variables is particularly low (McClelland & Judd, 1993). Latent Variables In this case one latent variable interacts with another latent variable. This is the most complicated case. Kenny and Judd (1984) have developed a solution using product indicators of X1M1, X1M2, X2M1, and X2M2, but it quite complicated with many nonlinear constraints and it requires a large sample size to have sufficient power and the assumption of normality to identify the model. Klein and Moosbrugger (2000) have developed a method of estimation that does not require nonlinear constraints and their procedure is described by Marsh, Wen, and Hau (2004). This method uses paired product indicators (X1 with M1 and X2 with M2).

30

Other Issues Three additional issues that are discussed here briefly are repeated measures, multilevel modeling, meta-analysis, moderated mediation or mediated moderation, and mixture modeling. Repeated Measures All of the above discussion presumes that the design is between participants. In some cases, the design is repeated measures. Judd, Kenny, and McClelland (2001) describe moderator analyses in this case. In essence, moderation is indicated by computing a difference score across conditions and determining whether the moderator predicts that difference: Because the difference score measures the effect of X on Y for each person, using it as the outcome variable gives an ideographic measure the causal effect and it is then determined if the moderator predicts that causal effect. Moderation with repeated measures can also be handled by multilevel modeling. Multilevel Modeling In some situations the data are said to be clustered, and a multilevel model is needed to model the nonindependence due to clustering. For instance, there might be students in classrooms with students being a level 1 and classrooms at level 2. Sometimes, there are level 1 moderators, these being moderators that vary within the classroom. More typically there are level 2 moderators, these being moderators that vary between classrooms. Note too that X can be at either level 1 or level 2. If X is measured at level 1, one can determine a generic moderator, that is, measure the extent to which there is variation in the X-Y relationship. Evidence of generic moderation would be obtained if there was variation in the X-Y slopes.

31

Meta-analysis Much of meta-analysis involves the study of moderation. If a variable predicts effect sizes, that variable is moderator. Moreover, as with multilevel modeling, one can test for a generic moderator by determining if effect sizes vary more than would be expected by sampling error. One of the key tasks in meta-analysis is the understanding or what are the moderators of the effect. Mediated Moderation and Moderated Mediation In mediated moderation, the moderation disappears when the mediator is introduced. In moderated mediation, the pattern of mediation varies as a function of the moderator. See my mediation page for more information. Papers by Muller, Judd, and Yzerbyt (2005) and Edwards and Lambert (2007) discuss the relationship between mediated moderation and moderated mediation. They also present examples of each. Also Preacher, Rucker, and Hayes have developed a macro for estimating moderated mediation (click here). Mixture Modeling We can use mixture modeling to search for a "latent" moderator. In such a case, we measure X and Y and then we allow for latent classes which would be the moderator variable.

Bibliography Aguinis, H., Beaty, J. C., Boik, R. J., & Pierce, C. A. (2005). Effect size and power in assessing moderating effects of categorical variables using multiple regression: A 30-year review. Journal of Applied Psychology, 90, 94-107. Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions . Newbury Park, CA: Sage. Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ: Erlbaum. Edwards, J. R. (1995). Alternatives to difference scores as dependent variables in the study of congruence in organizational research. Organizational Behavior and Human Decision Processes, 64, 307-324. Edwards, J. R., & Lambert L. S. (2007). Methods for integrating moderation and mediation: A general analytical framework using moderated path analysis. Psychological Methods, 12, 1-22.

32

Frazier, P. A., Tix, A. P. & Barron, K. E. (2004). Testing moderator and mediator effects in counseling psychology research. Journal of Counseling Psychology, 51, 115-134. Hamaker, E. L., Grasman, R. P. P. P., & Kamphuis, J. H. (2010). Regime-switching models to study psychological processes. In P. C. M. Molenaar & K. M. Newell (Eds.), Individual pathways of change: Statistical models for analyzing learning and development , 155-168. Washington, DC: American Psychological Association. Hayes, A. F., & Matthes, J. (2009). Computational procedures for probing interactions in OLS and logistic regression: SPSS and SAS implementations. Behavior Research Methods, 41, 924936. Judd, C. M., & Kenny, D. A. (2010). Data analysis. In D. Gilbert, S. T. Fiske, G. Lindzey (Eds.), The handbook of social psychology(5th ed., Vol. 1, pp. 115-139), New York: Wiley. Judd, C. M., Kenny, D. A., & McClelland, G. H. (2001). Estimating and testing mediation and moderation in within-participant designs. Psychological Methods, 6, 115-134. Kenny, D. A., & Judd, C. M. (1984). Estimating the nonlinear and interactive effects of latent variables. Psychological Bulletin, 96, 201-210. Klein, A. G., & Moosbrugger, H. (2000). Maximum likelihood estimation of latent interaction effects with the LMS method. Psychometrika, 65, 457-474. Kraemer, H. C., Stice, E., Kazdin, A., Offord, D., & Kupfer, D. (2001). How do risk factors work together? Moderators, mediators, independent, overlapping, and proxy risk factors. American Journal of Psychiatry, 158, 848-856. Kraemer H. C., Wilson G. T., Fairburn C. G., & Agras W. S. (2002). Mediators and moderators of treatment effects in randomized clinical trials. Archives of General Psychiatry, 59, 877-883. Marsh, H. W., Hau, K. T., Wen, Z., Nagengast, B., & Morin, A. J. S. (2011). Moderation. In Little, T. D. (Ed.), Oxford handbook of quantitative methods. New York: Oxford University Press. Marsh, H. W., Wen, Z. L., & Hau, K. T. (2004). Structural equation models of latent interactions: Evaluation of alternative estimation strategies and indicator construction. Psychological Methods, 9, 275-300. McClelland, G. H., & Judd, C. M. (1993). Statistical difficulties of detecting interactions and moderator effects. Psychological Bulletin, 114, 376-390. Muller, D., Judd, C. M., & Yzerbyt, V. Y. (2005). When moderation is mediated and mediation is moderated. Journal of Personality and Social Psychology, 89, 852-863.

33

- QTBRUploaded bySundas48
- !!! Methodology Mediation Moderation EmsleyUploaded byEliza DN
- med_moderation_analysisUploaded byzabishah007
- Mediating moderationUploaded byShahid Rashid
- Linear ModelsUploaded byelmoreilly
- Transformational Leadership and Work EngagementUploaded byNatalia Rios
- Century National BanksUploaded byAsad Msa
- MediationUploaded byHemant Kanuga
- 1-s2.0-S1877042813036136-mainUploaded byZefry Wahyu Purnama
- Simmons DissertationUploaded bysimranarora2007
- MacKinnon-Introduction to Statistical Mediation AnalysisUploaded byMartin Garro
- Beyond Baron Kenny Hayes2009Uploaded byhafiz10041976
- Beta Just DialUploaded byVivekSingh
- Two-Stage Stochastic and Deterministic OptimizationUploaded byshoemakerlevy9
- The Relationship Between Religious Attitudes and Perceptions of Organizational Justice in CoachesUploaded bythibashini
- Statistics Lecture 2Uploaded byLennox McGrain
- Additive modelsUploaded byhubik38
- Corporate social responsibilityUploaded byctnas1109
- Week 4 JeopardyUploaded byramatsree
- Business Statistics 1Uploaded bySabyasachi Mukerji
- 30120130406005Uploaded byIAEME Publication
- Evaluating the Impact of Health Card Program on Access to Reproductive Health Services: An Indonesian ExperienceUploaded byErlangga Agustino Landiyanto
- Learning by Thinking_Harvard Business School.pdfUploaded byOriginalo Versiona
- Teaching Statistical Thinking_ Part 1 Descriptive Statistics - Duke University _ CourseraUploaded byAndrea Vukic
- Linguistic description moderates the evaluations of counterstereotypical people.Uploaded byMark Rubin
- Mod20 (1)Uploaded bysuwarno haji
- understanding earnings quality - a review of their proxies and determinant and their consequencies (2010).pdfUploaded byFlorentina Widita
- Final Mid Tear Project QuestionnerUploaded byaksasheikh
- hasil01.docxUploaded byRisal
- 4thQ -5thLessonUploaded byJenny Eyen Nicole

- Pakistan Affairs CSS past papersUploaded byabdulhadiqureshi
- LR Insider TradingUploaded byAsif Khan Niazi
- ProductionUploaded byAsif Khan Niazi
- Chap 017Uploaded bysiefbadawy1
- Highlights ES 201314Uploaded byMike Kalo
- Management SciencesUploaded byAfaq Ahmed Soomro
- report AIOU.docUploaded byAsif Khan Niazi
- Asif Final CvUploaded byAsif Khan Niazi
- AsmUploaded byAsif Khan Niazi
- Aqt Development of Questionnaire iUploaded byAsif Khan Niazi
- Farasat 1 History of Economics in PakistanUploaded bySalman Ali
- Lecture 7 Questionnaires Interviews & ObservationUploaded byAsif Khan Niazi
- Lecture-8-Writing and Presenting Project ReportUploaded byAsif Khan Niazi
- Cointegration ResultsUploaded byAsif Khan Niazi

- 10. Demand ForecastingUploaded bygeorge
- BUSN603_11th_PPT_Ch_05Uploaded bycpd
- 1-s2.0-S0167587713000391-main.pdfUploaded byThefirefox
- admn_ma (2)Uploaded byKushal Agrawal
- Comparing SlopesUploaded byZefry Wahyu Purnama
- Chi Square HandoutUploaded byNavRattan Goyal
- Stat 200 Week 6 HomeworkUploaded byldlewis
- Jeff 18Uploaded bylarabellono
- week01Uploaded byRichard Ding
- intro statUploaded bykm3197
- MIT18_05S14_Prac_Fnal_ExmUploaded byIslamSharaf
- Solutions Chapter 9Uploaded byreloadedmemory
- Lambdin (2012) - Significance tests as sorceryUploaded byRB.ARG
- Bham13P3slidesUploaded byliemuel
- Krajewski Ism Ch13 solutionUploaded byalex_sasc
- Fallacies of Statistical SignificanceUploaded bydayalume
- 10. IJAMSS - Modify Bayes Estimation With Extension Loos Function for Parameter Weibull Distribution - Hadeel Salim Al KutubiUploaded byiaset123
- Stats lecture 1Uploaded byGSS
- Help Nonlinear RegressionUploaded bybennyferguson
- Basic Anova ConceptsUploaded byzenflesh
- Anova -All AboutUploaded byBrook Rene Johnson
- Inferential controlUploaded byMelissa Yeo Mixia
- Chapter 12Uploaded byJannat Khan
- QM SyllabusUploaded byUtsav Ghulati
- SylUploaded byfriendboy
- non parametric test of interaction.pdfUploaded byAgus Dwi Sulistyono
- syllabusUploaded by2fast4lyf
- SSRN-id1660104.pdfUploaded byEugenio Martinez
- STAT400Uploaded bysophie
- Spatial regretion: the curious case of negative spatial dependenceUploaded byJuan Pinzon

## Much more than documents.

Discover everything Scribd has to offer, including books and audiobooks from major publishers.

Cancel anytime.