Work motivation and social communication among public managers

Abstract This article tests hypotheses about the effects of two types of work motivation (i.e. intrinsic motivation and extrinsic motivation) and four types of social communication on three important work dispositions (i.e. job involvement, red tape, and perceived organizational effectiveness) among 790 managers employed in public agencies in the states of Illinois and Georgia. The results show that intrinsic motivation is significantly associated with public managers' job involvement, perceptions of red tape, and organizational effectiveness. Also, certain types of organizational communication and mentoring socialization are closely related to the outcomes. The results of cross-interaction effects between motivation and communications on outcome variables support the SDT prediction that when public managers are more involved with different types of social communication behaviors, the relationship between extrinsic motivation and three outcome variables becomes more pronounced. That is, communication in public agencies provides more autonomy and discretion for extrinsically motivated public managers, thus leading to more positive organizational dispositions. Keywords: job involvementmentoringorganizational behaviororganizational dispositionspublic human resource managementred tape Notes 1. An earlier version of this manuscript was presented at the 68th Annual Conference of the Academy of Management (AOM), Anaheim, California, held from 8 to 13 August 2008. 2. As teams, committees, and other groupings make up the overarching structure and culture of organizations, organizational communications travel through various channels and processes (e.g. team-based management and team-based/shared leadership). These different dimensions of communications such as horizontal, bottom-up, or local communications and hierarchical, top-down, or vertical communications occur in organizations. Local and group-based communication activities influence organizational outcomes. The data include no subunit-level data and hence we could not devise a three-level HLM analysis (i.e. including individual level variables, subunit-level variables, and organizational-level variables altogether). However, we emphasize that this study utilized the National Administrative Studies Project (NASP) III survey instrument, which collected data on perceptions and attitudes reported by managers who are working in public agencies in the states of Illinois (N = 358) and Georgia (N = 432). That is, the respondents represent the upper-level values, issues, and identities from a manager's perspective. 3. While we put a heavy emphasis on the positive side of organizational communication in this study, we also acknowledge the contrasting findings and arguments that organizational communications, such as mentoring, and other types of communications might impose structural, cultural, or communicative barriers that might have detrimental effects. For example, 'horizontal communications encounter difficulties as a result of conflict, competition, or other differences between subunits and groups' while 'vertical communications encounter difficulties as a result of hierarchical filtering and superior–subordinate relationships, including resistance, inattentiveness, misunderstanding, and reticence or withholding of information by persons at lower levels' (Rainey, p. 367). Lack of feedback, noise in communication, and misuses of language might also lead to ineffective communications and bring difficulties in organizations. Specifically, some authors point out that 'communication distortions' might be frequently observed in the public sector, which include the notions of 'jargon, inflated prose, and the manipulation of information for political or bureaucratic purposes' (Rainey, p. 368). 4. Although integrated regulation (i.e. autonomous motivation) is still regarded as extrinsic motivation, it has a full range of autonomous regulations and an internal locus of causality, characteristics very similar to those of intrinsic motivation (i.e. inherently autonomous motivation). In this stage, extrinsic motivation shares some qualities with intrinsic or autonomous motivation and is characterized as 'internalization of extrinsically motivated activities' with respect to particular behaviors (Gagne and Deci Citation2005, pp. 335–336). Intrinsic motivation and integrated extrinsic motivation were identified as the two distinct types of autonomous motivation. 5. The sample actually included some respondents classified as 'professionals', in addition to managers. Professionals often have levels of responsibility comparable to those of managers, and have experience with their agencies' administrative and personnel management systems, obviously. More importantly, various analyses indicate that the distinction between manager and professional does not relate significantly to the findings reported here. Results are available from the authors. 6. Although list-wise deletion (where cases with a missing score on any variable are excluded from all analyses and the effective sample size includes only cases with complete records) is the most common method for handling missing data, this approach sacrifices a large amount of data by eliminating all cases with any missing data (Roth Citation1994). The EM and the MCMC methods allow us to obtain an effective sample size and to minimize possible bias in parameter estimates in an HLM analysis. These approaches also fully capture the asymptotic properties of the underlying population. 7. A simple random sample of 2000 public managers was selected from original population (1000 from Illinois and 1000 from Georgia). An overall response rate of 42.47% was achieved (790 usable completed surveys out of 1853 in the original sample). Extensive documentation about the NASP-III (e.g. sampling procedures and survey instruments) is available at the project Web site, http://www.uga.edu/padp/nasp.html 8. Skewness can show whether the item's distribution deviates from the symmetry distribution. We can argue that skewness values outside the range of ± 2 would be problematic because this is a serious level of skew. Kurtosis measures the degree to which the area in a distribution is in the middle and the tails of a distribution. As a rule of thumb, the range of ± 2 is often considered as a significant departure from normality (Pedhazur Citation1997). In this research, most items show a relatively stable and similar amount of variance. The results of Kolmogorov-Smirnov and Shapiro normality tests suggest that items in this study do differentiate responses fairly well. In terms of individual normality, most items have a high positive kurtosis value, which means that most respondents have selected the same response option. Most variables of skewness or kurtosis are all between − 2 < s(k) < 2, and we can argue that these variables are approximately normally distributed. Relative multivariate kurtosis (1.155 < 2.0) also indicates approximate multivariate normality. 9. The alpha values indicate that the scale content is homogeneous and answers are consistent. Internal consistency is important in this research because the homogeneity of motivation and other scales is critical for measuring the respondents' attitudes about the survey items accurately. Moreover, when the content is consistent, it is also easier to interpret. To assess internal consistency and to ensure reliability of each scale, Cronbach's alpha reliability test was done. 10. Since EFA is an atheoretical method that might sometimes result in poor indexing, we also employed the CFA method, which is based on SDT, to develop more theory-oriented intrinsic and extrinsic motivation scales in this study. The table with factor loadings of EFA and CFA models is available from the authors. We also statistically tested the coefficient difference between intrinsic motivation and extrinsic motivation. In order to test for statistically significant differences between the two coefficients on outcome variables, we used an independent samples t-test method by dividing the sample on the basis of the two groups – intrinsic motivation (group 1) and extrinsic motivation (group 2). The result of the test showed that the effects of intrinsic motivation and extrinsic motivation on three outcome variables are statistically different and the difference is significant. 11. For example, the comparative fit indices (CFI) are more than 0.95 (greater than 0.90 is acceptable), and the root mean square error approximation (RMSEA) is 0.065 (less than 0.08 is acceptable). 12. Among these variables, 'gender' (female = 0) and 'ethnicity' (non-White = 0) were recoded as dummy variables. We also created ordinal-level variables (four categories) for 'job tenure (log)' and 'managerial power (log)' because responses to these items were highly skewed on the right or left side. 13. In measuring these constructs of motivation as well as other ordinal scale variables, 11 non-continuous ordinal variables (e.g. four-point Likert scale) were used, which might cause potential problems of nonlinearity and non-normality. Both non-normality and nonlinearity will generally result in underestimation of the relationship among variables. In other words, variable communalities, percentage of variance accounted for, and factor loadings will be lower than continuous and normally distributed data. As one solution to this problem, a polychoric-based solution was used in this model and polychoric correlation (PC) matrix, ranging from − 1.0 to 1.0, was developed for the measurement. 14. The level-2 variables were measured by the average scores on the variables for the respondents from each agency. That is, we aggregated individual data into level 2. The main rationale to aggregate individual data into level 2 in this research is social communication (i.e. organizational communication and mentoring socialization) usually occur within an organizational context and hence we needed to analyze those behaviors from an organizational (not individual) behavior perspective. In this regard, we examined level-1 and level-2 variables separately (even though all variables are individual level measures) and we aggregated the individual level data in organizational level (level 2). 15. Our measure of the communication variables measures the intensity (a percentage measure) of communication rather than measuring the actual quantity (a count measure) of communication. These communication indicators measure the relative amount of time that a manager spends communicating internally, relative to externally with business, or externally with government. Future research should use quantity or count measures to examine how 'intensity' of social communication would affect work motivation and organizational effectiveness. Nevertheless, the results in Table 3 indicate that relatively high levels of these three types of communication interact with extrinsic motivation. This supports our conclusions about the joint effects of communication and extrinsic motivation. 16. In this ANCOVA model, each slope of the covariate is assumed to have the same effect on each level of the factor (i.e. homogeneity of regression). Each predictor (covariate) was sequentially added and only one variable was retained when it shows that it has reliability greater than 0.05 and has a statistically significant random effect. Based on the results, eight predictors were retained at the employee level. 17. In this model, both of the un-centering (for dummy variables) and grand-mean centering methods were used because there is no random effect in the slope (that is, β1j is fixed across agencies). In other words, the level-1 covariates (X variables) were included to control for their effects on the outcomes, rather than to model between group variance on the slope of these variables. The intercept here is interpreted as the expected value of four outcome variables for each employee with an average score on each of the level-1 predictors. In this regard, grand-mean centering adjusts the variation in the intercept between agencies to control for differences in the level-1 predictors across agencies. 18. In the intercepts- and slopes-as-outcomes model, we can include level-2 predictors. After we have established that significant variation existed between group intercepts and slopes, we then use those parameter estimates as level-2 outcomes. The level-2 models are used for purpose of modeling the associations between level-2 predictors and the outcomes and to track the degree to which those predictors explain variance in the level-2 outcomes, In this model, a grand-mean centering option was used and variables were added by one at a time for model building while examining their coefficient for significance (of random effect) and reliability. 19. In the cross-level interaction model, an interaction is when the association between level-1 predictors and the outcome variables depend on the level of level-2 predictors. The cross-level interaction will be between the level-1 random effect and the level-2 predictors in the slope outcome model. 20. That is, mentoring programs in state agencies positively and significantly increase the mean levels of job involvement and perceived organizational effectiveness (i.e. grand mean of the outcome variables) by 0.347 and 0.694, respectively, whereas mentoring programs significantly lower perceived organizational red tape by 0.261. 21. Before moving forward to the analysis, we conducted a preliminary analysis to test the seriousness of the common method bias. One limitation of this research is that both independent and dependent variables come from a single data source, the NASP III manager survey. By running a principal component factor analysis for all variables included in the model, Harman's single-factor test indicates the seriousness of the bias: one can claim the bias is serious when only one factor is retained and it explains most of the covariance. The principal component factor analysis indicates that several factors are retained and the biggest factor explains 34% of the covariance, which implies that the bias is not very serious in this case. In addition, in order to minimize the common source bias, we measured the data at the two different levels – individual and organizational levels. For example, demographic controls and motivation variables were measured at the individual level (level 1) and organizational communication and mentoring variables were estimated at the organizational level (level 2). That is, although still using data from a same source, we changed the unit of analysis and tried to minimize the bias using HLM which can take into account the interdependence of individual-level observations nested within higher-level (i.e. agency-level) state agencies; hence, we can correct estimates of standard errors using HLM.