The Impact of Management Control on Employee Motivation and Performance in the Public Sector

Abstract

This study examines the relations among various types of management control, intrinsic and extrinsic motivation, and performance in the public sector. We draw on motivation crowding theory and self-determination theory to argue that four different types of management control (i.e. personnel, cultural, action, and results control) are likely to have an influence on intrinsic motivation and/or extrinsic motivation. We test a structural equation model using survey data from 105 similar departments in the public sector. Our findings indicate that the use of personnel and cultural controls is positively associated with employees’ intrinsic motivation, and that the use of results controls is positively associated with employees’ extrinsic motivation. Moreover, both intrinsic motivation and extrinsic motivation are positively associated with performance. Taken together, these findings support the idea advocated by New Public Management proponents that results control can enhance employee motivation and performance in the public sector. However, the findings also highlight an essential nuance; in addition to results control, personnel and cultural controls are also important, as they enhance intrinsic motivation and performance. This implies that a sole focus on results control is too narrow and can lead to suboptimal levels of employee motivation and performance in the public sector.

Keywords:

• Management Control
• Motivation
• Object-of-Control
• Performance

JEL classifications:

• H83
• M10
• M12
• M40
• M50

Acknowledgements

Earlier versions of this paper were presented at seminars at the University of Groningen, Catolica Lisbon School of Business and Economics, Alliance Manchester Business School, IE Business School Madrid, Erasmus University Rotterdam, University of Sydney Business School, and Ghent University and at the 39th European Accounting Association conference in Maastricht. We thank the participants for their feedback. We also would like to thank Sally Widener and the two anonymous reviewers for their constructive feedback on earlier versions of this paper. This paper is based on a chapter from a doctoral dissertation completed at the University of Groningen (van der Kolk, 2016).

ORCID

Berend van der Kolk http://orcid.org/0000-0003-2011-4585

Paula M.G. van Veen-Dirks http://orcid.org/0000-0003-4017-717X

Henk J. ter Bogt http://orcid.org/0000-0001-8016-2269

Notes

1 SDT makes further distinctions within the categories of extrinsic and intrinsic motivation on the basis of the degree to which motivation is autonomous versus controlled; extrinsic motivation is divided into ‘external regulation’ and ‘introjected regulation’, while intrinsic motivation is divided in ‘identified regulation’ and ‘integrated regulation’ (Gagné & Deci, 2005). Furthermore, other studies also further split intrinsic motivation in 'enjoyment-based' and 'obligation-based' intrinsic motivation (cf. Frey et al., 2013). In our paper the focus is not on such sub dimensions of intrinsic and extrinsic motivation, but on the relations between different types of MC and intrinsic/extrinsic motivation.

2 Various accounting scholars have argued that the use and usability of MC elements is context dependent (e.g. Merchant, 1982; Ouchi, 1979). This has resulted in a vast literature that considered environmental contingencies and the usability of certain MC elements, which ultimately would affect organizational performance (see e.g. Chenhall, 2003; Luft & Shields, 2003). Influential in this respect is a paper by Merchant (1982), which states that two factors together determine to a large extent the feasibility of different types of control: “knowledge of desirable actions and the ability to measure results on the important performance dimensions” (Merchant, 1982, p. 47). Merchant (1982) proposes that when the knowledge of desirable actions is excellent and when the ability to measure results on the important performance dimensions is high, the use of action controls and results controls is recommended to achieve good performance, while the use of personnel or cultural controls is not particularly recommended or discouraged in such a situation (later editions of the textbooks by Merchant and van der Stede (2007) identified ‘cultural’ controls as a separate category, but in Merchant (1982), ‘cultural’ control was still included in the broad category of ‘personnel’ controls). Although it is not the primary purpose of this research - which is to examine the relation between MC and motivation - we also include potential direct effects between MC and performance in our analyses.

3 The corresponding original texts can be found on the following pages of the Merchant and van der Stede (2007) textbook: results controls (pp. 29–32), action controls (pp. 76–79), cultural controls (pp. 85–90) and personnel controls (pp. 83–85).

4 An assumption that underlies formative constructs is that the items in the construct should cover all the dimensions of the construct, and that when two items relate to the same dimension, they should be weighted so that this dimension is not disproportionally represented in the final construct. We think of the five items per category as five separate dimensions as they all refer to a different part of MC that is not necessarily related to the other items in the category (it is, for instance possible to use rewards without using sanctions, although one might also argue that they are both part of the higher-order category 'incentives'). It could also possible, however, to argue that the items PRS(a) and PRS(c) are part of the same dimension ‘training’ within the construct of personnel control, and that the items RES(c) and RES(d) are part of the same dimension ‘incentives’ within the construct of results control (see Appendix A). We therefore also conducted a robustness analysis with weighted averages for these dimensions, and found that all hypotheses could also be confirmed when doing the analysis using the weighted averages.

5 We also conducted the analyses that are presented in the findings section with the original two-item construct by Gagné et al. (2010); all confirmed hypotheses from the trimmed base model could be confirmed except for hypothesis 4b (p = .131, two-tailed).

6 The number of municipalities is decreasing year by year, mainly because municipalities are ‘merging’ in order to be better able to deal with new tasks that are decentralized to municipalities and the (presumed) ‘economies of scale’. To illustrate this decline in the number of municipalities: in the year 1910 there were 1,121 municipalities, in the year 2000 the number had decreased to 537 and in the year 2010 there were only 431 municipalities. These numbers were retrieved from the Dutch Central Bureau for Statistics, http://www.cbs.nl/, on 23 July 2015.

7 In Dutch: Nederlandse Vereniging Voor Burgerzaken (NVVB), www.nvvb.nl.

8 2,918 were official municipal e-mail accounts, other email addresses (846) were ‘unofficial’ Hotmail or Gmail addresses, or were even linked to other, non-municipal organizations such as consulting firms, who were also subscribed to DAPA’s online newsletter.

9 We have multiple respondents (two, three, four or five) from 27 organizations and one respondent for the remaining 78 organizations. The number of municipalities with multiple respondents is low, which does not allow robust statistical analyses to check for interrater reliability. We visually inspected the data and we can verify that there was consensus among employees within municipalities about the use of different categories of MC elements. For instance, the average standard deviation for the MC-related items on the surveys from those working in the same municipality was significantly lower than the average standard deviations for the full sample.

10 In three cases, we inserted a computed value on the basis of our sample for municipal size (employees) and in one case for the variable departmental size (employees), because respondents filled in '0'. We used the relation between the variable that was missing and size (inhabitants) to insert a value, referred to by Kline (2011, p. 58) as "regression-based imputation". This is a relatively advanced technique that is accepted for imputing missing values when preparing for conducting structural equation modeling. Using regression-based imputation was necessary in order to be able to produce a full correlation table for all 105 observations.

11 This value for the Chi-square test is not surprising given our sample size, see Hair et al. (2006, p. 753).

12 To determine the full model including all the original items in a structured equation model without manifest variables would mean that 34 additional parameters have to be estimated. Because the rule of thumb for similar studies is that about 5–10 observations are necessary for every parameter that is estimated, this would mean that we would need at least 16 (the original relations and the covariances between the control types) + 34 (additional parameters) = 50 * 5 (minimum for rule of thumb) = 250 observations, which we do not have in the current study. A way to deal with this is using the manifest variables (summated scores for the constructs). A disadvantage of this method is that some of the variance of the original survey data is lost. However, the method allows us to estimate a path model, also for studies with smaller sample sizes (Kline, 2011, pp. 8–9, 11–12; O’Rourke & Hatcher, 2013, p. 9; Schreiber et al., 2006, p. 326; Widener, 2007).

13 We do not hypothesize or include a direct relation between INTM and EXTM in our base model (although including it would not affect our findings. We make this choice because the literature on which we draw states extrinsic controls may simultaneously enhance EXTM and reduce INTM, but the literature does not explicitly state, to the best of our knowledge, that a direct relation exists between the two types of motivation should be expected. Including a covariance between EXTM and INTM does not alter our findings, but because the first SEM model should be in line with what can be expected on the basis of the theory (Hair et al., 2006; Kline, 2011), we chose to report the base model excluding a direct effect from one type of motivation on another.

14 Fte stands for full time equivalent. These raw size variables had to be transformed because of nonnormality (Hair et al., 2006, p. 176), after which their skewness and kurtosis values were at the acceptable values (see Appendix B).