Leadership and Public Service Motivation in U.S. Federal Agencies

ABSTRACT

This analysis of over 6,900 federal employees’ responses to the Merit Principles Survey 2000 examines the influences of leadership and motivational variables, and especially public service motivation, on the “outcome” variables job satisfaction, perceived performance, quality of work, and turnover intentions. CFA confirms a factor structure for transformation-oriented leadership (TOL), public service-oriented motivation (PSOM), transaction-oriented leadership (TSOL), and extrinsically oriented motivation (EOM). Multivariate regression analysis shows that TOL and PSOM, as well as interaction effects of TOL-TSOL and TOL-PSOM, have strong relations to the outcome variables. SEM analysis examines direct and indirect effects of the main variables. Overall, the results indicate that TOL and PSOM have more positive relations to the outcome variables than do TSOL and EOM. The combination of high TOL and high PSOM has the strongest positive, and hence desirable, relation with organizational outcomes. Among this very large sample of federal employees, those who perceived their leader as displaying TOL (i.e., leadership that is encouraging, supportive, informative, and that emphasizes high standards) also expressed higher levels of PSOM and higher levels of job satisfaction, perceived performance and work quality, and lower turnover intentions. The SEM analysis further indicates that TOL has these effects by way of empowerment, goal clarification, and PSOM, and is distinct from TSOL (transaction-oriented) leadership, which shows no such relationships.

Notes

The numbers in parentheses are Cronbach's Alpha values.

*Correlation is significant at the 0.05-level (two-tailed).

**Correlation is significant at the 0.01 level (two-tailed).

^a Missing variables were recalculated and adjusted by the EM algorithm.

• Critical values are 1.96 for P < .05 and 1.65 for P < .10 (t-statistics are in parentheses).

• 22 Agency-Dummies (reference group: Other Agencies) were included as control variables in the OLS model. A listing of federal agencies is available from the authors.

*P < .10: significant at .10-level.

**P < .05: significant at 0.05-level.

^a Total effects can be calculated by summing up direct, indirect, and spurious effects.

•Not all total effects were included in this table. The full information about total and indirect effects is available from the authors.

**Values are significant at p < .01 and p < .05 (one-tailed).

The original version of this essay was awarded the Sage Publications Best Doctoral Student Conference Paper from the Public and Non-Profit (PNP) Division at the 2005 Annual Conference of the Academy of Management, August 5–10, in Honolulu, Hawaii.

For instance, Bass and his colleagues have revealed that the more effective leaders are both transformational and transactional (Hater and Bass 1988), and transformational and transactional forms of leadership are distinct but not mutually exclusive processes (Bass 1985a).

In the public sector, we can hypothesize that these two leadership types are sometimes combined and overlapped, which engenders interaction effects—the effect of transformational leadership on outcome variables is different, depending on the effect of transactional leadership. Based on this rationale, an interaction variable of TOL-TSOL was included in this study.

In this study, it is assumed that the relationship between these two types of work motivation is independent, additive (i.e., non-spurious), or even interactive. For instance, a lot of empirical evidence suggests that extrinsic or monetary incentives undermine intrinsic motivation either marginally or markedly, depending on other moderating organizational factors.

More broadly, PSM can be characterized as a reliance on intrinsic rewards (i.e., sense of accomplishment and of fulfilling a duty as a public employee) over extrinsic rewards (i.e., a pay raise, a promotion, job security, and pay for performance ratings) (Crewson 1997).

For example, according to Herzberg, the opposite of job satisfaction is not dissatisfaction but rather a simple lack of satisfaction; also, the opposite meaning of dissatisfaction is not satisfaction but rather no satisfaction.

Some of these were about more intrinsic, altruistic, and public service-oriented rewards, while others asked about extrinsic and more self-interested rewards such as a pay raise and a promotion. Many of these items asked respondents to choose three of these categorical items that will most motivate them.

For example, we can think of the situation where many employees are not only motivated by publicness itself, but also inspired by transformation-oriented leaders, which could have more positive and significant effects on job satisfaction and performance. Following this rationale, a TOL-PSOM interaction variable was added in this research.

Empowerment means giving employees the authority, skills, and self-control to perform their tasks. Although empowerment and managerial flexibility have similar constructs in their items, they are distinctive in that empowerment is more of a behavior-oriented variable whereas managerial flexibility is more related to the organizational structure and design issues. Based on this rationale, we separated these two variables, and factor loadings also statistically supported our distinctions between the two variables.

These variables were included to control and moderate the leadership and work motivation effects among federal employees, leading to more rigorous and accurate research findings. Moreover, by including these variables, we can observe whether the different types of leadership and work motivation will show the direct and indirect effects on outcome variables through these mediators in a structural equation model (SEM). In SEM, procedural equity perceptions and objective performance appraisal systems are included as moderators whereas goal clarity and empowerment function as mediators.

In this study “performance and quality of work” were operationalized by perceptual and subjective measures (rather than using objective ones) based on federal survey data. Previous empirical studies suggested that “there is evidence of a high correlation between perceptual and objective measures at the organizational level” and found that “measures of perceived organizational performance were correlated positively to objective measures of organizational performance” (e.g., see Kim 2005, 250).

The EM method utilizes an iterative method to impute missing values. This method consists of two steps and the process is iterated until the difference between the reproduced covariance matrices obtained by two adjacent iterations falls below some prespecified criterion (Little and Rubin 1987).

Using the VIF and R² test, we checked the multicollinearity in the independent and interactive variables. The tolerance level is about .3 and VIF value is less than 3, showing that no serious problems were found in this model, which could be due to using factor scores in regression analysis. Generally, using factor scores as explanatory variables is supposed to reduce multicollinearity problems.

One of the assumptions of factor analysis is interval data; however, Kim and Muller (1978) suggest that ordinal data can be used if it is regarded that the ordinal categories to the data do not seriously distort the underlying metric scaling. In the same vein, they argue that use of dummy variable data can also be allowed if the underlying metric correlations between the variables are thought to be moderate (.7) or lower. In this study, all these conditions are met.

After obtaining factor scores and, subsequently, new variables, we tested the reliability, or the internal consistency, using Cronbach's alpha. All the scales have an Alpha value of .7 or above.

The formula for factor scores is F jk = ∑WjiZik (F = individual factor scores; W = weighted values; Z = the standardized variables). We can have three advantages by using factor scores in regression. First, we can reduce or eliminate multicollinearity because the variables causing the multicollinearity will combine to form a factor. Second, using the factor index, we can make interval variables instead of ordinal or nominal variables because all ordinal level data can be transformed into interval data that have factor scores rather than 5-level Likert scales. Third, we can reduce the number of variables by making new variables.

In measuring these constructs of motivation, eight dichotomous variables were used, which might cause potential problems of nonlinearity and nonnormality—both nonnormality 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 polychroic-based solution was used in this model. Polychoric correlations (PC), ranging from − 1.0 to 1.0, were developed for ordinal or dichotomous data.

Theoretically, polychromic correlations should yield higher correlations among categorized variables, as they disattenuate for the effects of categorization. They should also result in higher communalities, percentage of variance accounted for, and factor loadings. Moreover, polychromic-based analysis yields a much clearer solution, with clear separation between the two factors.

Education level, job experience, and current pay grade are important as control variables especially for turnover intentions because turnover intentions could naturally occur without any effect. We can expect that these moderators can reduce the internal validity threat, called history effect.

Please contact the authors for a detailed explanation of the rationale for this turnover intention measure.

In ML estimation, the weight matrix is the inverse of the reproduced covariance matrix. The ML method is generally both scale free and scale invariant. It also assumes multivariate normality and, hence, non-normality would influence the significant test and the chi-square value. From multivariate normality tests, severe non-normality patterns are not observed and we can expect that this method would be more unbiased, consistent, and efficient, especially when the population distribution for the endogenous variables is multivariate normal (Kline 2005).

However, of the seven tests, the maximum likelihood chi-square test was inconsistent with a good model fit (χ² = 1590.73; p < .01). This particular fit index, which assesses the discrepancy between the sample covariance matrix and the implied covariance matrix based on the hypothesized model, is sensitive to sample size, with larger samples increasing the chi square and decreasing the likelihood of achieving a good model fit (James, Mulaik, and Brett 1982). Consequently, with large samples, virtually all models would be rejected as statistically untenable regardless of a good model fit (Kemery, Bedeian, Mossholder, and Touliatos 1985).

We also conducted an independent samples test to determine if there is a mean difference in four outcomes for PSOM, EOM, TOL, and TSOL. The results indicate that the mean level of four outcome variables is significantly different at .05 level for PSOM and EOM (e.g., in job satisfaction, t = 2.447, p = .015) and TOL and TSOL (e.g., in job satisfaction, t = 3.543, p = .008). Based on these results, we rejected H0 states and confirmed that the impacts of PSOM, EOM, TOL, and TSOL on outcome variables are statistically different.

In this regard, one of the rationales for adopting a structural equation model (SEM) in this research is to examine the indirect and spurious effects among antecedent variables and outcome variables, (e.g., turnover intentions), which cannot be measured by the OLS regression method.

For example, the “empowerment” variable gives a significant effect on performance and turnover intentions. Also, the effect of “objective performance appraisal systems” is highly and positively related to job satisfaction, performance, and quality of work.

Modification indexes (MI) show the amount by which the chi-square value would decrease if the suggested paths or error covariances were added to the model. In order to drop out chi-square values, we added a set of error covariance among indicators of latent variables (e.g., public service motivated-job satisfaction and empowerment-job satisfaction) as MI suggested.

Additional information

Notes on contributors

Sung Min Park

Sung Min Park ([email protected]) is Visiting Assistant Professor in the Department of Public Administration at University of Nevada, Las Vegas (UNLV). He was awarded the Sage Publications Best Doctoral Conference Paper from the Public and Nonprofit Division at the 2005 annual conference of the Academy of Management, and the Collins Award for best doctoral student paper from SECoPA in 2006. His primary research interests are public management, public human resource management, and quantitative research methods.

Hal G. Rainey

Hal G. Rainey ([email protected]) is Alumni Foundation Distinguished Professor in the Department of Public Administration and Policy of the School of Public and International Affairs at the University of Georgia. His research concentrates on organizations and management in government.