IMPLEMENTING SIMPLE SLOPE TECHNIQUES TO INTERACTION OF WORK VARIABLES

The purpose of this study is to investigate the effects of different levels of Organizational Identification (OI) and Education variables as moderators on the relationship between Percived External Prestige (PEP) and Organizational Commitment (OC). Study carried out among 206 white collor employees who were working automotive industry firms in Bursa/Turkey. In related literature some researchers displayed that PEP association with OC by moderetor role of Organizational Identification. Thus we accepted this model and investigate the impact of diffrent level of moderator variable(s) on focal predictor .For this purpose, we used best subset regression procedure and simple slope tecniques for identify the different levels effects of moderator variables. The results showed that Education and OI were not only basic moderators but also their different levels have produced remarkable and various impacts on PEP and OC relationship.


Introduction
The type and degree of attachment that individuals manifest toward their employing organizations is a topic that continuous to be of interest to researchers and practitioners alike. Thus it is important for managers and academicans to find new approaches for hold talent individuals inside the organizations. In this framwork recent researches based on social identity theory mentioned that favorable reputation or prestige perception among employees fosters positive attitude toward organization and attached individuals to organizations (Mael and Ashforth 1992;Dutton and Duckerich 1991;Dutton et. al, 1994;Smitdt et al 2001). Hence researchers assumed that employees' reputation perception about working organization will be new intangible assets of organizations for establishing emotional bonds between employers and their working organizations. In this context, this study examines the interaction between employees' prestige perceptions about working organization and their attachment to it. This research also presents new perspective to researchers, to analyse different level effects of moderator(s) on models for understanding variables interaction deeply which is considered as the limitation of behavioral science.

Literature Review
Perceived external prestige (PEP) defined as degree of organizational prestige when compared organization with other relates (Mael and Ashforth 1992). Concept also revealed as individual level interpretation and evaluation of organizational prestige based on employee's own information. (Bergami and Bagozzi, 2000;Smithd's et al., 2001). On the other hand organizational attachment was defined as an individual's psychological and behavioral involvement in a social group or unit of which he or she exist as a member (Tsui et. al 1992:554) In this framework Organizational identification was "perception of oneness with or belongingness to some human aggregation, (Mael and Ashforth, 1992) process of incorporating the perception of oneself as a member of a particular organization into one's general self-definition (Dutton et. al., 1994;Pratt, 1998;Herrbach, 2006) where commitment defined as, employee's emotional to, identification with and involvement in the organization thus individuals who commit their organization based on affective tone, remains in organization because they "want" to stay. (Meyer et al., 1990;Meyer and Allen 1991;Meyer et. al., 1993;Meyer and Allen 1997). Recent studies suggested that people's group-based status or prestige judgments have an impact on both their feelings about themselves and their behaviors toward their group (Ellemers, 1993;Tyler and Blader, 2002;Riordan et al., 1997;Oliver and Mignonac 2004). Based on Social Identity Theory (SIT) assumptions individuals tend to looking for positive social identity and self image for social approval. (Turner et. al 1979;Dutton et. al, 1994) Thus when members beliefs that outsiders see their organization in a positive light, organizations become more attractive for them and they proud to be part of and being a member of it (Cialdini et.al., 1976 Researchers also mentined that affective commitment was tends to be stronger in more positively evaluated groups based on these groups contribute more to a positive social identity (Ellemers, 1993;Ellemers et al., 1999: 373). Boezemon and Ellemers found that pride and respect from the organization predicts organizational commitment among volunteers. Also Carmeli and Freund cited that PEP and organizational commitment are related under concept of organizational effectiveness (Carmeli and Freund, 2002: 61-62;Freund, 2006;78-79) where Mayer and Schorman results noted direct relationship between value commitment and organizational prestige (Mayer and Schoorman, 1998). Consequenly result empirical and theorical determinations indicated that PEP has significant and positive effects on individuals organizational commitment (Herrbach et.al, 2004;Carmeli and Freund, 2002;Carmeli, 2005a;Carmeli, 2005b;Freund, 2006).
Based on theoritical assumptions and empirical findings cited above, this research was investigated the interactions effects between employees' prestige perception, organizational commitment and identification. In related literature although there has been limited research have been focus on variables interaction, some researches were found that identification and commitment effected differently by PEP, group formation (Selfselected/assigned group membership) and group size (Ellemers et al., 1999:372), and they noted that PEP was associated affective commitment with moderator role of organizational identification (Bergami/Bagozzi, 2000:570).Similar model have been found by Carmelli and his friends,where results also cited that demographic variables such as education level have statistically significant effect on variables interaction ( Carmelli et. al., 2006: 100). However those researches only cited that organizational identification has moderating role between PEP and organizational commitment relation as a cognitive component of multible identification conceptualization of Tajfel and Turner (1979). And they did not explain different level effect of organizational identification as moderator on PEP and organizational commitment relationship. On the other hand the different level effect of moderator variables would be produced differences between focal variable impacts on dependent variable. Thus researcher must be take into consider of this issue in to their studies. Limitation of related behavioral literature that did not take into consider of deeper analysis on effects of moderator variable on focal predictor, this study was investigated effects of different levels of moderator variable on the relationship between PEP and organizational commitment. In this framework we used best subset regression analysis to get best equation among all possible regression models and used simple-slope techniques in order to determine the effect of different levels of moderators on indepented variable.

Participants
The data used in this study taken from automotive industry firms in Turkey which were stand first on the Bursa Chamber of Commerce and Industry's annual ranking of Bursa's Most Admired 250 Companies list on 2007. Questionnaires send to 400 white collar employees who were worked as a manager and 206 usable questionnaires was received. Participants' 83 percent were male, 51.2 percent held a B.A degree and 12.6 percent an MA degree. Respondents' organizational tenure range from 1 to 16 year, with 23.2 percent between 4-7 year, 37.7 percent between 8 and 15 year, 18.4 percent were over 16 year. Both of them are full-time employees and 75 percent of participants are married.

Measures
Organizational Commitment: Affective commitment to organization was assessed with the six-item affective commitment instrument which was developed by Meyer et. al (1993)  Perceived External Prestige: This measure is based on Fortune magazine's Annual Survey of "American's Most Admired Corporations" Index 8 attributes an measure has been used by numerous scholars, including Fombrun and Shanley (1990), Fryxell and Wang (1994), Carmelli (2002Carmelli ( ,2005aCarmelli ( , 2005bCarmelli ( ,2006Carmelli ( ,2004. The eight attributes were lined up as quality of management, quality of product, innovativeness, long-term investment value, financial soundness, develop and retain talent people, community and environmental responsibility and use of corporate assets. For this study overall index has been used but we divided "develop and retain talent people" attribute to two component part for avoiding misunderstanding. We asked respondents to assess their firm HRM policies by "My Company has a reputation among its key competitors for having better investment to his members" and "My Company has a reputation among its key competitors for having high level of employee quality". We have found using factor analysis nine items were loaded on a single factor. (α =0.86; GFI= 0.94; AGFI= 0.89; CFI=0.95; RMSR=0.04; RMSEA: 0.06) Control variables: The respondents were asked to indicate their age, sex, marital status and education level. Those demographic variables were used to control the relationship between the independent variable and the dependent variable.

Methodology
In this paper a two-step procedure has been used in order to analyses the data set. In first stage, we used all possible regression procedure to add the most appropriate demographical factors to regression equation (1). After determining the best subset regressors, we estimated the model with full interaction terms and used simple slope and simple intercept procedures in order to deeper analyses of moderation effects.
Following the above-mentioned literature we assume, In order to add most appropriate demographical factors (age, education, martial status, sex) to equation (1), all possible regression procedure has been used. In that process, PEP, OI and their interaction term (OIxPEP) treated as fixed predictors for all models, and intercept term, age, education, martial status and sex treated as free predictors. Finding an appropriate subset of regressors for the model is called the variable selection problem because this process involves two conflicting objectives. Researchers want to use as many regressors as possible so that the information content in these factors can influence the predicted value of dependent variable on the other hand researcher want to use as few regressors as possible because the variance of the prediction increases as the number of regressors increases (Montgomery and Peck 1992). All possible regression procedure is a computational technique for variable selection and it requires estimating all the regresion equations involving all possible subsets of the pool of potential predictors and identifying for detailed examination a few good subsets according to some criterion (Neter et al., 1996).
We used three criteria for evaluating subset regression models. Our first criteria is the Adjusted 2 R statistic which allow us to avoid the problems of interpreting 2 R . We choose the model that has a maximum adjusted  (Montgomery and Peck 1992). We choose the model, which has a minimum residual mean square. Our third criteria is Mallow's p C statistic which allow us to determine the regression equation with little bias (Rencher, 1995). We choose the model that has a minimum p C statistic. After determining the best subset regressors, we estimated the regression equation with full interaction terms and used simple slope and simple intercept procedures. An interaction occurs when the magnitude of the effect of one independent variable on a dependent variable varies as a function of other independent variable(s) (Preacher et. all, 2006). This is also known as a moderation effect. Simple slope and simple intercept procedures are allow us to understand the nature of the conditional relation (Akien and West, 1991). If the interaction term is found to be significant at a given significance level, the regression of dependent variable on a focal predictor is typically probed across values of the moderator(s).
Let us, following equation is determined as best subset regression at the end of all possible regression procedure.
here, if 7  is statistically significantly different from zero, and taking into consideration the literature behind equation (1) These four equations makes it clear that, by using simple slope and simple intercept procedures, moderator effect can be examined deeply. Additionally, these four simple slope and simple intercept are statistically different from each other because the coefficient of interaction term 7  in equation (2) is statistically significantly different from zero. In other words, testing 7  in equation (2) statistically different from zero is equivalent to testing these four simple slope and simple intercept are statistically different from each other. On the other hand, testing each one of the simple slope is statistically different from zero, can be tested by a t test but in this case, standard errors of the simple slopes must be determined as follows,

Factor Analysıs
In order to get factor scores of scales (Organzational Commitment, Organzational Identification, Percieved External Prestige), we applied confirmatory foctor analysis and explanotary factor analysis to our data set. Following Table1-3 show GFI, AGFI, CFI, RMSEA,  and factor loadings of items. According to results of Factor Analysis, each of the uniqe scales loaded on a single factor. So that, as mentioned methodology section, we assumed the valid model which is accepted in related literature that Organizational Identification has mediating role between Percieved External Prestige and Organzational Commitment is as followig, 1 2 In order to determine this model's efficiency, we also consider to add demographical factors to equation (1). For this purpose, based on the methodology section, we used all possible regression procedure. Here by results of analysis in table 4.

Alphanumeric Journal
The Journal of Operations Research, Statistics, Econometrics and Management Information Systems ISSN 2148-2225 httt://www.alphanumericjournal.com/  Table 5 shows estimation results of Equation (2). Following Preacher et. all (2006), one can claimed that interaction occurs when the magnitude of the effect of one independent variable on a dependent variable varies as a function of other independent variable(s), this is also known as a moderation effect. In our case, interactions occur when the magnitude of the effect of PEP on OC varies as a function of OI and EDUC. As can be seen in Table 5, all coefficients are statistically significant at %5 significance level in our final model. Moreover, model has adj. 2 R tatistic 0.51 which is a considerable explanatory level. If we compare this result with Adj. 2 R statistic of Model 1 which is presented in Table 4. our final model's variance explanation power is increased %2. Final model's Covariance matrix can be seen in Table 6.

Simple Slope And Simple Intercept Analysis
Estimation results of equation 2 and variance covariance matrix can be seen on Table 5 and VI respectively..
By using this equation (3) and by choosing specific values of moderators we investigated the magnitude of the effect of PEP on OC varies as a function of OI and EDUC. EDUC is a dichotomous variable and OI is continuous variable therefore, we used 0 and 1 for EDUC and one standard deviation below from mean and one standard deviation above from mean values of OI are used as specific values. This process allows us to examine the moderators effect extensively. By using those specific values of moderators in equation (3), we reached the following sub-equations, which were; allow us to understand the impacts of low and high levels of moderators on the relationship between PEP and OC.
These four sub-equations can be shown in Table 7 below, According to table VII, When PEP is zero, the mean value of OC for the group, which includes individuals who are Low educated and have Low OI scores, is equal to 1.51. If Education and OI are fixed in those low levels, every 1 unit increases in PEP will causes 0.14 unit increases in OC scores (see Model No1 in table 7.).

Alphanumeric Journal
The Journal of Operations Research, Statistics, Econometrics and Management Information Systems ISSN 2148-2225 httt://www.alphanumericjournal.com/ When PEP is zero, the mean value of OC for the group, which includes individuals who are Low educated and have High OI scores, is equal to 2.37 and every 1 unit increases in PEP will causes 0.14 unit increases in OC scores for that group (see , model no 2 in table 7).
When PEP is zero, the mean value of OC for the group, which includes individuals who are High educated and have low OI scores, is equal to 1.48 and every 1 unit increases in PEP will causes 0.35 unit increases in OC scores for that group (see, model no 3 in table VII).
When PEP is zero, the mean value of OC for the group, which includes individuals who are High educated and have High OI scores, is equal to 1.95 and every 1 unit increases in PEP will causes 0.67 unit increases in OC scores for that group (see, model no 4 in table VII).

Discussion
In thes sub section we compared four different groups which can be shown on Table 7. If one compares Model no 1 and Model no 2 he can claim that while education is fixed in its low level, a change in OI scores from their low level to high-level causes 0.86 unit increases when PEP is equal to zero. Although there is statistically significantly differences between the slope parameters of the models no1 and no2 in table VII, differences of the impact of slopes is so close (0.02). So, a change in OI scores from its low level to its high-level, when education set low level, causes very small effect to magnitude of the effect of PEP on OC. If ecucation is low the direct effect of OI on OC is remarkable, the moderating effect on PEP would be neglected.
While education is fixed in its high level, a change in OI scores from its low level to its high-level causes 0.47 unit increases when PEP is equal to zero. In that case, every one unit increases in PEP will cause additional 0.32 unit increases in OC scores. So when education is high, a change in OI scores from their low level to highlevel,the magnitude of the effect of PEP on OC drastically increases (see model no. 3 and 4). Thus if education is high; not only direct effect of OI on OC but also moderating effect of OI on PEP dramatically increases. Hence, we could say that, high educated employees were more carry weight to prestige perception that low educated ones.
On the other hand, OI scores are fixed their low level (models no1 and no3), a change in education from its low level to its high-level causes statistically significant but small decreases (0.03) when PEP is equal to zero. In that case, every one unit increases in PEP will cause additional 0.21 increases in OC scores in. Therefore, when OI scores of individuals are low, if education level moves from low to high, the magnitude of the effect of PEP on OC drastically increases. Thus, one could say that, when set OI in low level, direct effect of education on OC is not considerable. However, while moving education low to high, its produced conspicuous increases on moderating effect on the relationship between PEP and OC.
While OI scores are fixed in its high level, a change in education from its low level to its high-level causes again decreases (but this time considerable amount, 0.42 units) when PEP is equal to zero. In that case, every one unit increases in PEP will cause additional 0.55 unit increases in OC scores (see, model no2 and no4). When OI is fixed in its high level, a change in education from its low level to its high-level causes drastically increases on the magnitude of the effect of PEP on OC.

Conclusion
In this research. we accepted PEP association with OC by moderetor role of Organizational Identification model and investigate the impact of diffrent level of moderator variable(s) on focal predictor. Then we used best subset regression procedure and simple slope tecniques for identify the different levels effects of moderator variables. This process allows us to examine the moderators effect extensively. The results showed different