Group-Based Incentives and Individual Performance: A Study of the Effort Response

Abstract

Group-based incentives are attractive in contexts where production is interdependent. Prior work shows such incentives increase group performance despite freeriding concerns, yet little is known about the effort response of individuals. Using individual-level data, the authors assess the introduction of group-based performance pay using difference-in-difference estimation. Overall, performance increased by 19%. Nearly all workers contributed to this effect. Further, two-thirds of this effect stems from increased efficiency (more output per unit of time) and one-third from higher attendance. Both incentive and selection effects are present. By leveraging individual-level data, the authors pose new questions and evidence to the group-based incentives literature.

Keywords

difference-in-difference performance pay group-based incentives freeriding incentive effects selection effects absenteeism efficiency performance productivity

When group-based performance incentives are introduced to a production process at a firm, how do workers respond? While such incentives align with the optimization needs of interdependent production processes when one worker’s output is another worker’s input (Boning, Ichniowski, and Shaw 2007), as economists, we know to be concerned about freeriding (see Kandel and Lazear 1992). Group-based incentives may not be sufficiently strong to induce efficient effort by an individual worker given that the performance-contingent payment is divided between group members, yet the full cost of effort is borne by the individual (Alchian and Demsetz 1972; Holmstrom 1982). Therefore, introducing group-based performance incentives will not necessarily induce an optimal effort response by individual workers.

Despite this freeriding concern, prior work has shown that introducing a group-based performance incentive can increase productivity of the group (Knez and Simester 2001; Hamilton, Nickerson, and Owan 2003; Boning et al. 2007; Friebel, Heinz, Krueger, and Zubanov 2017). Yet, what about the individual effort response of the worker? Does the introduction of such incentives affect all workers equally, or does the effort response vary across workers? Setting freeriding aside, we might expect that low performers respond more as they have more room for improvement (Hansen 1997). Further, for workers who increase their effort, do they do so by showing up to work more often, or by being more productive when working (or both)?

This article is uniquely poised to answer these questions. We do this by analyzing the effects of introducing a group-based performance pay incentive in a firm that relies on a complex and interdependent production process. The company—Hydrema A/S—builds large construction machines (dump trucks, backhoe loaders, excavators, and so on) at two (almost identical) factories, one located in Støvring (Denmark) and the other located 800 kilometers away in Weimar (Germany). Both factories used fixed pay through the second quarter of 2015, at which time the Støvring factory introduced a group-based performance pay system and Weimar remained on fixed pay. Note that years prior to the incentive adoption, Hydrema invested in a monitoring system in both Denmark and Germany to track plant performance by aggregating the performance of individual workers. The incentive system was explicitly designed as group-based because of the interdependent production process, yet individual performance data continued to be collected. This setup is ideal for our difference-in-difference analysis of the effects of adopting a group-based performance pay system on individual worker performance.

Main Contribution

Our main contribution is to shed light on individual responses to the introduction of a group-based performance incentive.¹ To this end, we conduct a comprehensive analysis of the Hydrema A/S company, when part of the company moved to group-based incentive pay, combining information from personnel records, interviews, financial data, and annual reports. These sources allow for a rich understanding of individual effort responses to group-based incentives in the context of a complex and interdependent production process.

We find an overall increase in performance of 19% after the Støvring plant introduced the group-based incentive.² While the overall positive effect on performance adds another finding to an increasingly dense literature on group-based incentives, more important in terms of our contribution is highlighting the individual worker’s effort response.³ By leveraging individual data on performance, we shed light on implicit assumptions in the group-based incentives literature and pose questions not broached in prior work. We view our study as having three important contributions.

First, we examine heterogeneity in effort response across workers, providing direct evidence about how a group-based incentive motivates workers across the distribution of performance. We estimate heterogeneous worker responses using a conditional quantile difference-in-difference approach (Koenker and Basset 1978) and the unconditional quantile regression approach (RIF-OLS) proposed by Firpo, Fortin, and Lemieux (2009). The results show that the introduction of the group-based incentive caused a large positive effort response throughout the performance distribution (with some exceptions at the extreme percentiles). Hence, the overall positive effect on performance induced by the adoption of the group-based incentive is a consequence of (nearly all) workers having higher performance.

This insight into who responds—nearly all in this case—has been largely absent from the group-based incentives literature to date. Key papers in this literature are limited to, at best, group-level outcomes (e.g., company-wide ranking in Knez and Simester 2001; team performance in Hamilton et al. 2003; overall production line performance in Boning et al. 2007, and store-level in Friebel et al. 2017), implying that conclusions about individual effort response are indirect (i.e., because group performance increased, at least some individuals responded) and the distribution of response is unknown.⁴ An interesting exception is Hansen (1997), using individual-level data from both before and after the introduction of group-based performance pay. His results show an improvement for initial “low performers” and no effect for initial “high performers”; yet the setting (call center) and performance measure (call time) limit the conclusions that can be drawn about individual performance response in complex production settings.⁵ In this article, we show with direct and reliable measures of performance that the individual worker response was essentially uniform.

Second, we separate the individual worker’s effort response into two components: efficiency (performance per unit of time) and attendance (showing up for a scheduled shift). We find that two-thirds is attributable to increased efficiency (intensive margin), and one-third follows from higher attendance (extensive margin). Prior literature on incentives, both group and individual, has paid little attention to these two effort dimensions separately; implicitly, performance increases in the literature have been assumed to be efficiency increases.⁶ Given these two margins of individual response, prior findings on the effects of group-based incentives on performance of the group may have over-attributed any gains to increased efficiency of the worker.

Third, we also decompose the main effect of 19% into selection and incentive effects. While this decomposition is well-known in the incentives literature, the evidence stems from individual-based incentives (Lazear 2000; Franceschelli, Galiani, and Gulmez 2010), not a group-based context. It is not obvious that prior work showing high-performers are attracted to firms with individual-based performance incentives (Lazear 1986, 2000) would apply to jobs with group-based incentives. Indeed, group-level pay seems on the surface more appealing to low performers. Our results, however, show that one-quarter of the performance gain follows from more productive workers being present at the firm under the group-based performance pay system (selection effect), and that three-quarters is attributable to individual workers becoming more productive (incentive effect). Under the original pay system, we observe that newcomers and leavers were of fairly similar performance quality, but after the introduction of the group-based incentive, newcomers were of significantly higher performance quality. Therefore, we find evidence of positive selection even for group-based incentives.

Overall, by leveraging individual-level performance data, we contribute new questions and evidence to the literature on group-based incentives. While these contributions have relevance to both academics and practitioners, an outcome most relevant to firms is whether the introduction was profitable, which requires that the value of the productivity gain exceed the costs associated with the change in pay system. To this end, back-of-the-envelope calculations based on annual reports and personnel records show that the firm transformed the performance improvements into favorable financial results. By combining insights on worker effort response and firm performance with robust identification, this study’s end-to-end coverage strengthens its appeal for studying (and teaching) about the effects of group-based incentives.

Setting

Background on the Company

Hydrema A/S is a production company making large construction machines. Out of the company’s 520 employees, 300 are blue-collar production workers. The company has production facilities in Støvring (Denmark) and Weimar (Germany). The company headquarters are located in Støvring, but otherwise the two production sites have identical setups; and their locations, 800 km apart, imply very little interaction between the two production sites, except for coordination at the management level. The blue-collar workers we study have no contact across workplaces, eliminating contamination concerns. This arrangement makes the Weimar plant an ideal control group for the changes we study in Støvring.

The production process at Hydrema is highly interdependent and, like most other plants producing complex products, the production process involves a very large set of components and tasks. A large fraction of parts, including steel plates, arrive at the factory as raw materials and are welded and machined before they are painted and sent to the assembly line. Between 2,000 and 2,500 components are put together before the final product (a large construction machine) leaves the factory. Most important, all workers contribute to the same production process, and the output of one worker in a given department is the input to another worker in the same or another department. Yet, within departments, production is dynamic with workers moving to a variety of tasks as needed (i.e., not a fixed assembly line). As such, a group-based performance pay system rather than individual-based performance pay aligns with the production setting.

Performance Measures

Important for the analysis is the measurement of performance. To this end, Hydrema uses a comprehensive computerized monitoring system (CMS) that was implemented years prior to the change in incentive system. All tasks in the complex production process are carefully described and given a takt time, which is the standard time it takes to complete the task. This takt time can be compared to actual work time (as a percentage basis), and the difference relative to 100% reflects worker efficiency on a given work order. These measures are recorded in the CMS.

Key for our analysis is that performance is measured at the worker level. The firm’s performance measure has three components. For the company to earn money, workers need to be present, they have to work efficiently, and they must be allocated to a productive work order.⁷ This implies that the performance measure used by the firm consists of the same three components: attendance, efficiency, and time on (productive) work orders. In practice, attendance $(A)$ is a simple measure of the proportion of time the employee is present at the workplace relative to planned work hours; efficiency $(E)$ measures the worker’s speed on a given work order relative to takt time; and the time on productive order $(P)$ measures the proportion of time the worker is on a value-creating work order. Hence, the firm operates with the overall performance measure:

Performance = A \times E \times P

(1)

The two dimensions the workers have agency over are A and E. By contrast, P requires managerial involvement. While Lazear (2000) explicitly focused on E in his landmark study of individual incentives, studies in the group-based incentives literature implicitly equate changes in performance for the group as improvements in efficiency. The full picture of worker response requires taking the product between efficiency and time at work. Failing to account for possible increases in time worked overstates efficiency gains.

Treatment

The Støvring plant moved from fixed pay to group-based performance pay after the second quarter of 2015. The fixed pay system was characterized by individual base pay $(w_{i})$ reflecting education, skills and experience, and a performance threshold $(q_{0})$ , for which performance below the threshold was considered incompatible with continued employment. The new performance pay system maintained the base pay and threshold but added a performance contingent bonus. This arrangement implied that worker compensation could not fall when transiting to the new system.⁸

The performance pay system rewards improvements in performance at the department (D) level (typically 30 workers) and plant (C) level with equal weight. Given the interrelated and dynamic production process, and to secure collaboration and coherence, it was important to the firm that the incentive system reward improvements at both group levels.⁹ To measure improvements in performance, benchmarks $(B_{j, k}, j = {A, P, E}, k = {D, C})$ for each of the performance components were calculated as averages over the year preceding the summer of 2015 for both the department and the plant level; these benchmarks remained fixed (i.e., not a moving target).

The bonus intensity parameters $(b_{j}, j = {A, P, E})$ vary across the three performance components. For attendance, an improvement of 1 percentage point is rewarded with an additional Danish krone (DKK) 0.21 per hour; productive time improvements increase the hourly wage by DKK 0.24; and efficiency improvements lead to an increase of DKK 0.22 per hour.¹⁰ It follows that bonus payments, with actual performance measures being $P_{j, k}, j = {A, P, E}, k = {D, C}$ , become:

{B o n u s}_{j k} = \sum_{j, k} [b_{j} (P_{j, k} - B_{j, k})], j = {A, P, E}, k = {D, C}

(2)

For worker i employed in department D under the group-based performance pay system pay is:

{P a y}_{i} = w_{i} + m a x {{B o n u s}_{j k}, 0}

(3)

While this shift from fixed pay to group-based performance pay seems straightforward, it is only the ultimate consequence of an involved process explicitly initiated by the CFO a few months earlier, and implicitly years in the making. He explains that, before the change, much time was spent on discussions at the yearly wage negotiations as to how to set pay. In those times, pay had no direct link to performance, despite a CMS system that produced reliable performance measures, which were used only to gauge plant performance and progress in production flows. In these discussions, the idea arose that these performance measures should be used as input in the pay system. Consequently, a committee comprising representatives from management, consultants from The Confederation of Danish Industries (employer organization), the unions, and workers was formed with the purpose of developing a performance contingent pay system.¹¹ This work started around March 2015 and took a few months. Ultimately, a local pay agreement containing the group-based performance pay system described above was signed. Hence, the treatment we are considering in the article is the shift from fixed-pay to a system of fixed-pay with the addition of a group-based incentive designed and implemented in committee.

The presence of the committee is important for the empirical analysis. Anticipation of a pay change may influence worker behavior and shift variables prior to the actual implementation. Reassuringly, this is not the case in our setting. Estimates (presented below) are unaltered when we shift the “treatment date” by one or two quarters, and placebo tests do not detect changes prior to the actual date of implementation (see Online Appendix 2).

Data and Descriptive Statistics

Our main analysis is based on comprehensive company-wide data for the period 2014q1 to 2018q2. The data set contains 3,525 worker-quarter observations: 1,676 originate from Støvring (treatment site) and 1,849 from Weimar (control site). In parts of the analysis, we apply a longer panel for Støvring, spanning 2009q1 to 2018q2 and containing 2,965 observations.

The main outcome variables are (overall) performance and the key performance indicators (KPIs) of attendance, efficiency, and time on productive orders. The intuition is that to create value for the company, workers have to be present (attendance), they have to be allocated to a productive work order, and they have to work effectively on a productive work order.

Table 1 shows the descriptive statistics for Støvring (the treatment site) for these variables for the longer panel of data. Average performance is 65.97, which is the product of attendance, efficiency, and time on productive orders. Attendance is generally high, with workers being present 95.95% of the time. The efficiency measure shows large heterogeneity: On average, workers perform at 85.78% of takt time, but the standard deviation of 31.65 gives rise to a notion of slow and speedy workers. The last measure shows that workers are on productive work orders 80% of the time on average. This measure has a relatively high standard deviation, reflecting that some workers almost always are allocated to productive work orders, while others spend more time on “unproductive” work orders such as cleaning, fixing defects and broken items, and so forth.

Table 1.

Performance and Key Performance Indicators for Støvring

	Sample period 2009q1–2018q2
	Performance	Attendance	Efficiency	Time on productive orders
Mean	65.97	95.95	85.78	80.06
Standard deviation	30.07	8.83	31.65	18.50
Median	64.54	100	85.36	85.95

Notes: Performance = Attendance x Efficiency x Time on productive orders. Number of observations: 2,965.

The first evidence on the treatment effect is presented in Figure 1. We show the average worker (overall) performance over time and across plants when data are available for both Støvring and Weimar (2014q1 through 2018q2). The combination of similar growth patterns in Støvring and Weimar prior to the treatment, and the divergence in average performance by plant upon introduction of the treatment presents compelling descriptive evidence of the incentive effect on overall worker performance.

Figure 1.

Average Worker Overall Performance across Time and Plant Location.

Further evidence is presented in Table 2. We present average worker performance by plant separately for both the pre- and post-periods. In the pre-period, the Støvring and Weimar locations are very similar with no statistically significant differences in average worker performance. In the post-period, average worker performance is significantly higher by 13.86 percentage points in Støvring relative to Weimar. These preliminary results show a large divergence in average worker (overall) performance after the introduction of the group-based incentives in Støvring.

Table 2.

Performance and Key Performance Indicators by Plant Location and Treatment Period

	Sample period 2014q1–2015q2 (pre-period)
	Støvring	Weimar	Difference
Overall performance	57.47 (22.94)	57.87 (24.31)	−0.40
Observations	561	615	1,176
	Sample period 2015q3–2018q2 (post-period)
Overall performance	83.33 (28.85)	69.47 (24.70)	13.86***
Observations	1,115	1,234	2,349

Notes: Averages (standard deviations) of worker overall performance.

p value < 0.10; **p value < 0.05; ***p value < 0.01.

We have previously argued that the Weimar plant is a close to ideal control for the Støvring plant, and we strengthen this argument in the following three ways. First, as stated in footnote 1, the CFO, who was instrumental in implementing the treatment, made it clear that the treatment affected only Støvring (Weimar was not part of the change). When asked if other changes occurred in Støvring at the same time, such as changes in the production process, the answer was a clear no.

Second, productivity growth in the manufacturing sectors in Denmark and Germany are highly correlated.¹² In the sample period 2014 to 2018, average growth in the manufacturing sector is 0.8 percentage points in both Denmark and Germany, and the correlation coefficient is 0.45; in the post-treatment period (2015–2018) the correlation is as high as 0.65. Hence, growth conditions in the two countries are similar both in the pre- and post-periods.

Third, above we established that pre-treatment performance levels are similar across locations (they differ by only an insignificant 0.40 percentage points).¹³ From Figure 1 it is also clear that pre-trends appear very similar. When we formally test the common trends assumption, the null of common trends cannot be rejected. That is, if we regress performance on a dummy for treatment group (Støvring), year dummies, and their interactions using only data from before the pay change, the joint test of the interactions is insignificant (F-stat of 0.29 and a p value of 0.917).

These findings in combination with the fact that we have one treatment, which is constant across groups and over time, allow us to estimate the treatment effect using conventional difference-in-difference methods (Chaisemartin and D’Haultfoeuille 2020), which we do in the next section.

Empirical Analysis

We estimate the effects of introducing group-based performance pay using a difference-in-difference approach. We establish a statistically significant increase in average overall performance of 19%. The mechanisms leading to this performance improvement are established in subsequent analysis. Existing workers have 14% higher performance in the new regime, and the remainder is attributable to selection. Hence, three-quarters of the overall performance increase follows from workers becoming more productive and one-quarter is the result of more productive workers being employed by the firm when the group-based incentive system is in operation. This result is robust and persistent, and it includes limited heterogeneity across workers. When decomposing the result, we find that the performance increase is primarily driven by higher worker efficiency, with some contribution from improved attendance.

Performance Effect

Our first estimates of how the introduction of group-based performance pay affects performance rely on comparison of performance in the pre-implementation period to performance in the post-implementation period (i.e., pre-post estimator) (Table 3, columns (1)–(4)). In the first regression, we use data from the longest panel available for Støvring (from 2009q1 to 2018q2). When we regress the natural log of performance on a dummy for the post-period (and controls), we obtain a significant effect of 0.46 (column (1)). When worker fixed effects are included, the coefficient is reduced to 0.11 but remains statistically significant (column (2)). Therefore, average overall performance is significantly higher in the post-period relative to the pre-period at Støvring. When we limit the time period to 2014q1–2018q2 (when data are also available for Weimar), we obtain similar results. The simple pre-post estimate is 0.40 (column (3)), and when worker fixed effects are included, we obtain an estimate of 0.10 (column (4)).

Table 3.

Performance Effects Using Difference-in-Difference Approach

	Støvring		Støvring		Støvring & Weimar		Støvring & Weimar
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	2009q1–2018q2		2014q1–2018q2		2014q1–2018q2		2014q1–2018q2
Post-period (PP)	0.46*** (0.04)	0.11*** (0.05)	0.40*** (0.06)	0.10* (0.05)	0.21*** (0.05)	0.01 (0.06)	0.45*** (0.03)	0.33** (0.15)
Støvring					0.03(0.07)	−0.55(0.39)	0.03**(0.01)	0.24(0.31)
PP × Støvring					0.18**(0.08)	0.14**(0.06)	0.19***(0.01)	0.14***(0.02)
Worker fixed effects	No	Yes	No	Yes	No	Yes	No	Yes
Quarter-year fixed effects	No	No	No	No	No	No	Yes	Yes
Observations	2,965	2,965	1,676	1,676	3,525	3,525	3,525	3,525
R-squared	0.17	0.57	0.18	0.64	0.11	0.61	0.13	0.62

Notes: All regressions control for a quadratic in tenure and dummies for the person being a new hire or separating. Clustering is at the plant by quarter-year level.

p value < 0.10; **p value < 0.05; ***p value < 0.01.

The difference-in-difference estimates are presented in Table 3, columns (5)–(8). In these models we use data from both Støvring and Weimar and regress the natural log of performance on a post-period dummy, a dummy for Støvring, an interaction between these two variables, and controls. In the specification with time (quarter by year) fixed effects, presented in columns (7) and (8), we obtain a significant treatment effect of 0.19 (column (7)); including worker fixed effects reduces the estimate to 0.14 (column (8)).¹⁴ This set of regressions constitutes the classical Lazear decomposition (Lazear 2000), which splits the overall performance response into incentive and selection effects. That is, of the 19 percentage point increase in performance for Støvring in the post-period relative to the pre-period, the incentive effect is 14 percentage points (or 74%), reflecting that existing workers show higher performance in the post-period. The remaining 5 percentage points in increased performance is the selection effect reflecting the greater proportion of higher performing workers in the post-period relative to the pre-period. Hence, the results show that the change in incentives significantly affected both the performance of incumbent workers and the selection of workers into and out of the firm. We will return to these points below when we explicitly look at worker selection and how incumbent workers become more productive.

The introduction of the group-based incentive was a process that took place over several months as noted above. Talks leading to the shift in pay regime started months before the actual implementation. To the extent that this process affected performance prior to treatment, our estimates would be biased. We approach the issue by blocking out time periods leading up to the shift from fixed pay to group-based performance pay. In the second column of Table 4 we exclude the quarter leading up to the change in pay system, and in the third column we exclude the six months prior.¹⁵ Irrespective of such restrictions, the results are robust and remain similar to the benchmark effects on performance (column (1)).

Table 4.

Sensitivity Analysis

	Benchmark	Exclude2015q2	Exclude2015q1 & 2015q2	1-year effect	2-year effect	Level regression
	(1)	(2)	(3)	(4)	(5)	(6)
	2014q1–2018q2			2014q1–2016q2	2014q1–2017q2	2014q1–2018q2
Post-period (PP)	0.33**(0.15)	0.34**(0.15)	0.28*(0.16)	0.61**(0.28)	0.34(0.23)	19.01***(6.41)
Støvring	0.24(0.31)	0.24(0.34)	-0.01(0.33)	3.49*(1.75)	1.35(0.98)	4.05(17.69)
PP × Støvring	0.14***(0.02)	0.14***(0.02)	0.15***(0.02)	0.12***(0.02)	0.13***(0.02)	11.92***(1.19)
Worker fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations	3,525	3,321	3,126	1,941	2,693	3,525
R-squared	0.62	0.63	0.63	0.71	0.68	0.66

Notes: The first model is the benchmark model from Table 3, column (8), presented here for comparison. All regressions control for a quadratic in tenure, dummies for the person being a new hire or separating, and quarter-year fixed effects. Clustering is at the plant by quarter-year level.

p value < 0.10; **p value < 0.05; ***p value < 0.01.

We also explore if worker responses to the shift in pay system were immediate (columns (4)–(5)), the concern being that a too long post-period could begin to pick up effects unrelated to the treatment. When limiting the post-period to one or two years, respectively, we obtain performance estimates that are statistically similar to the benchmark result. Yet, we observe a moderate increase in the estimate from 0.12 to 0.13 as we expand the analysis window from one year to two years after introduction. If this increase can be attributed to the performance pay system, then this indicates there may be learning over time in addition to the immediate effort response, which is a point we return to below. Overall, we find that these restrictions are inconsequential for our estimate of the treatment effect.

In a final regression (Table 4, column (6)) we estimate the model using the level of the performance variable instead of the natural log. We obtain a positive and significant treatment effect of 11.92. With a base of 65.97 (Table 1), this finding reflects an increase in performance of approximately 18%, which is comparable to the natural log results.

Heterogeneous Performance Effects?

The average treatment effect potentially shades underlying heterogeneity across workers, which is of particular importance for group-based incentives given freeriding concerns. Previous research has addressed heterogeneity in worker responses to incentives in a variety of settings with mixed results.¹⁶ Franceschelli et al. (2010) established fairly homogenous worker responses, but it is most common to find differences across workers (Hansen 1997; Hamilton et al. 2003; Burgess et al. 2017; Friebel et al. 2017).

We approach the issue of heterogeneity estimating worker responses at different percentiles in the performance distribution. We do this using two approaches: conditional quantile regression (QDID) following Koenker and Bassett (1978) and the unconditional quantile regression approach (RIF-OLS) proposed by Firpo et al. (2009). The QDID estimates are presented in the left panel of Figure 2. Results show that most workers have a positive response to the group-based incentive and the effect on performance is very similar from quantile 10 to quantile 90, with some deviations in the tails. Results for the RIF-OLS model presented in the right panel of Figure 2 show that for the lowest quantiles, the shift to group-based performance pay may have a zero or even negative effect, but for all quantiles above the 10th quantile the impact is positive. Hence, irrespective of the estimation method (conditionally or unconditionally), workers respond positively and (almost) uniformly to the new group-based incentive.¹⁷

Figure 2.

Heterogeneous Performance Effects Using QDID and RIF-OLS

What Is the Source of the Individual Performance Response?

The average increase in individual performance resulting from the adoption of the group-based performance pay is 14 percentage points. This increase has three potential sources. Workers can increase their attendance, they can become more efficient on work orders, and—manager willing—they can increase their time on productive work orders (as opposed to fixing defects and broken items or being idle). Effort response in attendance and efficiency reside solely in the worker (i.e., not managers).

We decompose the main effect into these three sources in Table 5 using the difference-in-difference approach. The results show an increase of 3 percentage points in attendance following the introduction of the group-based incentive. Worker responses in efficiency show a 10 percentage point increase. Performance gains due to more time on productive work orders are negligible. Hence, the primary source of increased performance is worker efficiency, with attendance as a secondary dimension.¹⁸ Decomposing these margins of effort response is a unique contribution of our article, going beyond the now common finding that group-level performance increases, on average, upon introduction of group-based incentives.

Table 5.

Decomposition of Performance Effect

	Attendance	Efficiency	Productive work orders
	(1)	(2)	(3)
Post-period (PP)	0.01(0.03)	0.44***(0.05)	−0.12(0.10)
Støvring	0.09(0.10)	0.79**(0.30)	−0.63*(0.37)
PP × Støvring	0.03***(0.01)	0.10***(0.01)	0.02(0.02)
Worker fixed effects	Yes	Yes	Yes
Observations	3,525	3,525	3,525
R-squared	0.29	0.66	0.59

Notes: All models have the same specification as column (8) in Table 3, except for the change of dependent variable. All regressions control for a quadratic in tenure, dummies for the person being a new hire or separating, and quarter-year fixed effects. Clustering is at the plant by quarter-year level.

p value < 0.10; **p value < 0.05; ***p value < 0.01.

Understanding Individual Effort Responses to Group-Based Incentives

Perhaps the most consequential finding from this article, given well-established freeriding concerns in the literature (Alchian and Demsetz 1972; Holmstrom 1982; Kandel and Lazear 1992), is the strong performance response by nearly all workers following the adoption of the group-based performance pay system. While it is well-established that group-based incentives can increase group-level performance, we know little about how response varies across individual workers. Hansen (1997) documented that initially low performers increased performance in response to group-based incentives, but high performers did not if we assume call length captures individual performance for call-center employees. Hamilton et al. (2003) found that when workers can volunteer for group work and are paid based on group performance, the performance effects for groups can be highly heterogeneous (i.e., variance of group performance greater than variance in individual performance), indicating that group incentives can amplify individual performance variation.

To What Extent Does the “How” Matter?

So, why did an across-the-board performance increase occur in this setting? An interview with the CFO, who was instrumental in the design and implementation process of the new incentive system, sheds light on the issue. First, he explains that the introduction of the new incentive system was preceded by the formation of a committee comprising management, union representatives, workers, and consultants who assessed ideas related to the new pay system. This committee worked to create the final incentive system that added a group-based variable pay component to the existing fixed pay system so that pay would not fall for workers. The CFO also noted that workers were initially skeptical of the pay change, but the attitude changed over time; using such a committee may therefore be critical to facilitating worker buy-in and trust.¹⁹

Second, as mentioned previously, the structure of the group-based incentive weights department- and establishment-wide performance equally. This approach was intentional given the interrelated production process (i.e., across workers and departments) and complexity that requires departments, at times, to allocate its most capable employees to the most challenging tasks. The equal weight implied that workers had to focus on their department’s performance with an eye toward the overall performance of the company.

Taken together, the design and implementation process are likely key contributors to the effort response and shifts attention in the literature toward questions of “how” in addition to “what.” In the words of the CFO: “I see the largest advantage [of the group-based incentive system] as having a common goal”; and he continues by sharing about a clear transition from an “us versus them” mentality for how workers and managers related to one another, to feelings of being “all in the same boat” with shared expectations and shared goals. Similar reasoning was used by Knez and Simester (2001) who argued that “paying all employees a bonus based on satisfaction of a common goal, Continental’s incentive scheme introduces externalities between the efforts of employees” (p. 764). These features of the implementation and design process are likely contributors to the widespread performance effects observed at Hydrema. By contrast, the introduction of a group-based incentive itself does not necessarily increase connections between employees. Indeed, despite their focus on the importance of social cohesion in understanding performance response, Delfgaauw et al. (2022) found no evidence that adopting a group incentive increases social cohesion of the group. Note, however, that if we contrast implementation—or the how—of the group incentive in Delfgaauw et al. (2022) to that studied here, the two starkly differ in terms of worker involvement.

Multiple Levers of Employee Response

This article also highlights the workers’ multidimensional effort response. We find increased effort in both efficiency (output per time unit) and presence (hours). To our knowledge, this decomposition is new to the incentives literature more broadly.²⁰ Prior work focuses on efficiency (i.e., output per unit of time) either explicitly (e.g., Lazear 2000) or implicitly. Ignoring the hours response will either underestimate worker effort responses or imply that conclusions on group-based incentives overattribute any performance effects to worker efficiency.

Furthermore, we contribute to a recurring theme in the incentives literature: worker sorting in response to changes in pay systems (Lazear 1986, 2000). Our results show a 19% increase in performance when shifting from fixed pay to group-based performance pay; increased worker performance (i.e., effort response) accounts for three-quarters of the effect and one-quarter is attributable to worker selection. This positive sorting in the context of group-based performance pay is intriguing.

To shed more light on what is causing positive selection, we take a careful look at the components driving the selection effect and start with churning patterns. The hiring rate in the Støvring plant is 5.2% in the pre-period and is significantly lower (2.7%) in the post-period (test for equality has p value = 0.009). Employee turnover also drops in the post-period: In the pre-period turnover was 6.6% and it falls to 3.6% in the post-period (test for equality has p value = 0.006). Finding significantly lower churn after adoption of performance pay differs from past findings (e.g., Lazear 2000).

The second component of the selection effect is the performance gap between leavers and newcomers. In the pre-period, newcomers performed on average 6.29 percentage points better than leavers, yet, in the post-period this performance gap increased to 12.49 percentage points. We illustrate this difference in Figure 3; we show the extracted fixed effects from log-performance fixed-effect models (similar to those presented in Table 3, column (8)) estimated separately for the pre-period and the post-period. The results confirm that the quality is more similar between leavers and newcomers in the pre-period, whereas newcomers have systematically higher fixed effects than leavers in the post-period.

Figure 3.

Performance Fixed Effects for Leavers and Newcomers

These findings imply that the positive selection effect established at Hydrema is driven by a large performance gap between leavers and newcomers in the post-period and comes about despite relatively low churn. We find it interesting that selection effects can also be established in low turnover settings like Hydrema and that it is not only found in high-turnover companies such as the one originally studied by Lazear (2000).

Finally, we assess the timing of the effort response. Was the effort response immediate or did it take time to manifest, and how durable was the effect? Using the post-treatment window comprising all years of data in Table 4 (column (1)) as the benchmark, we assess the time path of the effort response in Table 6 using a linear model (column (2)) and time-period dummies (i.e., each interacted with Støvring; column (3)). For the linear model, it shows a significant treatment effect for the duration of the data, suggesting a lasting effect. When we consider the effect of the treatment separately by time period, we see evidence that the response was both immediate and durable; further, the magnitude of the effect is largest in the final time period considered (i.e., 18+ months). These findings suggest that the effort response we observed is from workers becoming immediately more productive—perhaps they had knowledge on how to increase productivity yet were not sufficiently motivated to do so prior to treatment—as well as from workers learning ways to become more productive, which may take longer to achieve. Given that learning is a type of effort, this finding points to another dimension of response. Overall, these results suggest that following the incentive change, workers exerted effort both to increase performance immediately and to learn how to improve and sustain performance more permanently.

Table 6.

Learning Effects

	Performance
	Benchmark	Linear	Time Periods
	(1)	(2)	(3)
Post-period (PP) × Støvring	0.14***(0.02)
Months since PP × Støvring		0.02***(0.00)
1–6 months since PP × Støvring			0.13**(0.03)
7–12 months since PP × Støvring			0.10***(0.02)
13–18 months since PP × Støvring			0.13***(0.02)
18+ months since PP × Støvring			0.17***(0.02)
Worker fixed effects	Yes	Yes	Yes
Level-effects included	Yes	Yes	Yes
Observations	3,525	3,525	3,525
R-squared	0.62	0.62	0.62

Notes: All regressions control for a quadratic in tenure, dummies for the person being a new hire or separating, and quarter-year fixed effects. Clustering is at the plant by quarter-year level. For convenience, we present “Benchmark,” which is the same model as column (8) in Table 3. The “Linear” and “Time Periods” models estimate how performance develops in the period after group-based performance pay is introduced.

p value < 0.10; **p value < 0.05; ***p value < 0.01.

Conclusion

In response to the introduction of a group-based performance pay plan at a plant with complex, interdependent production, worker performance increased by 19 percentage points. The performance increase was across the board, with limited evidence of freeriding. When decomposing the effort response, three-quarters of the effect is brought about by increased performance of existing workers, while the remaining quarter is attributable to a greater proportion of high performers among newcomers following the change in pay system. Further, workers largely obtained the performance increases by improving their efficiency; increased presenteeism was a secondary, yet significant margin of response. Collectively these results provide new insights on how individual workers respond to group-based incentives, which is information that has been largely absent from the literature to date, with its focus on group-level outcomes. Hence, despite being a case study,²¹ the availability of individual-level data on performance enables us to ask new questions as well as provide fresh insights on existing questions in the incentives literature.

Alongside these contributions, two points should be highlighted. First, the treatment we study is the addition of a group-based incentive to fixed pay. Yet, we consider both the design (broad incentive system with equal weight of both department and company performance) and the implementation (committee work allowing for discussion between management and workers) as part of the treatment. Alternative treatments could be an “overnight” management decision to change pay (no committee) or implementation without a buy-in process (e.g., by excluding workers from the committee). Such alternative treatments would potentially yield results that differ from what we have found here. Insights from the treatment studied here point to the need for future work that focuses on the “how” of incentive implementation, particularly for group-based incentives given freeriding concerns.

Second, it is also natural to shine light on the German plant serving as a robust control site. This plant has not (yet) shifted to performance pay despite a likely positive outcome. In our conversations with the CFO, he assured us that it was because of limited time and external factors. This point underscores that implementation (the treatment) is more than just adopting any group-based performance pay at a random point in time; instead, successful implementation requires a careful and consultative process with multiple parties represented, including workers.

Our analysis details the individual effort response of the worker, adding new insights to the literature on group-based incentives. At the same time, a key question remains for the firm: Was it worth it? Was the adoption of group-based performance pay profitable for the firm? A profitable outcome for the firm would only be the case if the value of the increased worker performance exceeds the labor and other costs connected to the change in pay system. In Online Appendix 4 we address incentive system- and worker-related costs together with financial data. Our rough calculations show evidence of increased profitability for the firm. These calculations stress the complex nature of group-based performance pay; yet they show that it is possible to implement such a system in a profitable way.

In all, our work fills an important gap in the incentives literature by showcasing the individual effort response to group-based incentives. Direct evidence of a largely uniform effect on performance as well as evidence of positive selection had been missing from group-based incentive literature. In addition, this article highlights multiple effort margins for worker response (i.e., efficiency compared to hours), relevant for implementation of individual and group-based incentives. These results are obtained from a study explicitly addressing the design and implementation of the incentive. Future research may benefit from more careful attention to such aspects when studying the consequences of pay incentives.

Supplemental Material

sj-pdf-1-ilr-10.1177_00197939231220033 – Supplemental material for Group-Based Incentives and Individual Performance: A Study of the Effort Response

Supplemental material, sj-pdf-1-ilr-10.1177_00197939231220033 for Group-Based Incentives and Individual Performance: A Study of the Effort Response by Anders Frederiksen, Daniel Baltzer Schjødt Hansen and Colleen Flaherty Manchester in ILR Review

Footnotes

Acknowledgements

We thank Hydrema A/S and, in particular, CFO Johnny Larsen, for providing the data and for being available when we had questions. We are also grateful for the comments we received from seminar participants, in particular Tor Eriksson.

The article builds on early work by Daniel Baltzer Schjødt Hansen.

ORCID iD

Anders Frederiksen

Colleen Flaherty Manchester

An Online Appendix is available at .

For information regarding the data and/or computer programs used for this study, please address correspondence to Anders Frederiksen at afr@btech.au.dk.

1

Earlier research has identified that management practices may come in bundles (Ichniowski, Shaw, and Prennushi 1997; ). This bundling did not happen in the case of Hydrema A/S. In an interview with the CFO we asked: “Back in the summer of 2015, Hydrema chose to move from fixed pay to performance pay in Støvring but kept the fixed pay system in Weimar. Did you make any changes to the production process at that point in time, or was it only a change in the pay system?” The CFO responded: “It was only the pay system, which we changed. Process changes were not implemented in Hydrema in any way.” The follow-up question was: “And what about Weimar, did anything change at that point in time?” The answer was prompt: “No, Weimar was not part of this.”

2

An earlier estimate of the effect of introducing group-based incentives on production line performance is provided by . They showed that steel mill lines became more productive when group-based incentives were introduced, and when such incentives were implemented in conjunction with problem-solving teams, lines could gain up to 20% of unrealized yield.

3

Why was an individual performance pay incentive not adopted given individual-level worker data? An interview with the CFO and a plant visit made it clear this was off the table given that the production process is interdependent as well as dynamic (i.e., not a rigid assembly line), requiring the most skilled workers to tackle the most challenging tasks. Individual performance pay would restrict this dynamic response by incentivizing the most skilled workers to remain on simpler tasks in order to outperform and maximize an individual incentive payment, to the detriment of division and plant productivity.

4

Indeed, questions about freeriding remain a key interest in this literature; recent work strives to provide greater insight into the question of freeriding by examining heterogeneity with respect to social cohesion to store-level performance after the introduction of performance pay (Delfgaauw, Dur, Onemu, and Sol 2022), yet inferences about individual-effort response remain indirect.

5

Hansen’s study has several caveats, which are recognized by the author. The dependent variable is minutes per call, and as the author states, one should be careful in interpreting shorter call times as a good measure of improved performance. The reason is that calls can be transferred between workers and workers differ substantially in their methods in that some prefer to call back clients instead of leaving them on hold.

6

A recent paper by Miller, Petrie, and Segal (2019) had a focus similar to ours and used an experimental laboratory setting where subjects perform a real job to show that incentives alter individual responses at both the intensive and extensive margins.

7

We use the term “productive work order” when we refer to a value-creating activity. The alternative would be a work order, which is not value creating, such as cleaning, unplanned breaks, or fixing defects and broken items.

8

Note the similarity to the system studied in Lazear (2000). In Safelite, workers have a guaranteed wage if performance is low and receive individualized performance-based pay for higher levels of performance. This guarantee is intended to mitigate turnover among low performers, who would have left the company (been made worse off) if a full-blown pay-for-performance system were implemented. When a small, group-based incentive was introduced at Continental Airlines (Knez and Simester 2001) and at the bakery chain studied by , it was also as an “add-on” bonus. Similar considerations were made at Hydrema, where workers were concerned that they could lose money under the new system. These concerns were explicitly addressed by management and the CFO states that: “The signal we sent from the beginning was that this would never have a negative consequence for the workers: We will never take anything away—it will always be an add-on to the existing wage.” With this perspective of only upside risk/gain, the new system has the potential to increase the “value of the contract.” Thus, it is possible that some of the performance effects observed could be an “efficiency wage” response (i.e., workers put forth additional effort to retain their employment given improved contract relative to outside options). The fact that it is a group-based incentive makes this explanation more salient given that take-home pay could increase in response to just one worker being more productive. Regardless of the channel—an efficiency wage or a pay-for-performance explanation—introduction of the group-based incentive is poised to increase productivity.

9

When asked if individual performance pay was ever considered, the CFO replied that this option could have been discussed given the CMS, but it was never on the table. He continually referred to a need for flexibility and collaboration in the production process, which could be challenged if pay was based on individual-level (or even department-level) performance.

10

These numbers reflect that roughly one-third of the value of the performance gain goes to the workers. With an average salary of, say, DKK 170, a 1 percentage point increase in all KPIs leads to an hourly wage improvement of DKK 0.67 or 0.394%. The results below show an average performance increase of 19 percentage points; hence, on average, workers are paid 7.5% more after the introduction of performance pay.

11

also referred to discussions among managers and work councils prior to implementation of the bonus system, and that the implementation of the new pay system was made possible because of trust between the parties.

12

According to OECD statistics ().

13

We also have access to a limited set of worker characteristics (see ). Comparison across plants shows a common picture in that both plants have relatively high worker stability as average tenure levels are 7.64 years (SE = 6.52) and 9.50 years (SE = 5.08), respectively.

14

F-stat for the individual fixed effects is 10.867 (p value = 0.000). We explore the sensitivity of the difference-in-difference estimate and find robust results. Using quantile regression, the estimate for the median is 0.132 (SE = 0.027), and when we apply the changes-in-changes model by , we obtain a point estimate of 0.154 (SE = 0.068). We also explore the sensitivity to different levels of clustering. In the main regression we cluster at the plant by quarter-year level and obtain an SE = 0.018. Clustering at the department by quarter-year level produces an SE = 0.027, which is slightly larger than the presented results. Without clustering we obtain an SE = 0.028, clustering at the plant level leads to an SE = 0.004, and clustering at the individual level implies an SE = 0.050. An alternative is to use bootstrapping and in this case SE = 0.027. In all these cases the conclusion about statistical significance does not change.

15

We present placebo tests in . Such artificial treatment effects are small in magnitude and insignificant, showing no indication of worker effort responses in the quarters prior to implementation of the group-based incentive.

16

An attempt to estimate heterogeneous worker responses was also made by Franceschelli et al. (2010) in their study of a shift from fixed pay to individualized performance pay. They used pre-period data (i.e., the fixed-pay regime) to estimate individual fixed effects from performance regressions, and based on these they classified workers into ability groups. When doing so, they found similar performance responses across groups when the firm switched to performance pay. A series of other studies rely on interaction effects (Hansen 1997; Burgess, Propper, Ratto, and Tominey 2017; Friebel et al. 2017) and establish effect differences due to variation in initial productivity levels, location, and unit size. included illustrations that vividly show heterogeneity in worker responses, which is driven by timing and worker selection.

17

In we decompose the response into Attendance, Efficiency, and Time on a Productive work order. Each KPI shows a homogeneous response across the majority of workers.

18

Given high attendance rates (95.95%), the scope for improvement was limited. Were these increases in attendance beneficial, or were they at the expense of workers coming in ill? While we cannot determine exactly, we do not find evidence of a longer-term rebound or decline in attendance.

19

Trust may also have been facilitated by the fact that performance monitoring was already in place, although not yet used for pay; workers were accustomed to the measurement prior to the performance pay implementation.

20

Work by on this issue may also be instructive.

21

For practical matters, we asked the CFO: “Do you think there is something particular or unique about Hydrema, which made this a success. Something we cannot find in other companies?” He answered: “I don’t think so. We are not unique in any way. The most important part is that we had data. Well-known data that we could use in calculations. If you have such data [data that can guide the system] then I think it can be implemented in all companies.”

References

Alchian

Armen A.

Demsetz

Harold

. 1972. Production, information costs, and economic organization. American Economic Review 62(5):777–95.

Athey

Susan

Imbens

Guido W.

2006. Identification and inference in nonlinear difference-in-differences models. Econometrica 74(2):431–97.

Bloom

Nicholas

Van Reenen

John

. 2007. Measuring and explaining management practices across firms and countries. Quarterly Journal of Economics 122(4):1351–408.

Boning

Brent

Ichniowski

Casey

Shaw

Kathryn

. 2007. Opportunity counts: Teams and the effectiveness of production incentives. Journal of Labor Economics 25(4):613–50.

Burgess

Simon

Propper

Carol

Ratto

Marisa

Tominey

Emma

. 2017. Incentives in the public sector: Evidence from a government agency. The Economic Journal 127(October):F117–F141.

Chaisemartin

Clément de

D’Haultfoeuille

Xavier

. 2020. Two-way fixed effects estimators with heterogeneous treatment effects. American Economic Review 110(9):2964–996.

Delfgaauw

Josse

Dur

Robert

Onemu

Oke

Sol

Joeri

. 2022. Team incentives, social cohesion, and performance: A natural field experiment. Management Science 68(1):230–56.

Firpo

Sergio

Fortin

Nicole M.

Lemieux

Thomas

. 2009. Unconditional quantile regressions. Econometrica 77(3):953–73.

Franceschelli

Ignacio

Galiani

Sebastian

Gulmez

Eduardo

. 2010. Performance pay and productivity of low- and high-ability workers. Labour Economics 17(2):317–22.

10.

Friebel

Guido

Heinz

Matthias

Krueger

Miriam

Zubanov

Nikolay

. 2017. Team incentives and performance: Evidence from a retail chain. American Economic Review 107(8):2168–203.

11.

Hamilton

Barton H.

Nickerson

Jack A.

Owan

Hideo

. 2003. Team incentives and worker heterogeneity: An empirical analysis of the impact of teams on productivity and participation. Journal of Political Economy 111(3):465–97.

12.

Hansen

Daniel G.

1997. Worker performance and group incentives: A case study. Industrial and Labor Relations Review 51(1):37–49.

13.

Holmstrom

Bengt

. 1982. Moral hazard in teams. Bell Journal of Economics 13(2):324–40.

14.

Ichniowski

Casey

Shaw

Kathryn

Prennushi

Giovanna

. 1997. The effects of human resource management practices on productivity: A study of steel finishing lines. American Economic Review 87(3):291–313.

15.

Kandel

Eugene

Lazear

Edward P.

1992. Peer pressure and partnerships. Journal of Political Economy 100(4):801–17.

16.

Knez

Marc

Simester

Duncan

. 2001. Firm-wide incentives and mutual monitoring at Continental Airlines. Journal of Labor Economics 19(4):743–72.

17.

Koenker

Roger

Basset Jr.

Gilbert

Jr.

1978. Regression quantiles. Econometrica 46(1):33–50.

18.

Lazear

Edward P.

1986. Salaries and piece rates. Journal of Business 59(3):405–31.

19.

Lazear

Edward P.

2000. Performance pay and productivity. American Economic Review 90(5):1346–61.

20.

Miller

Amalia R.

Petrie

Ragan

Segal

Carmit

. 2019. Does workplace competition increase labor supply? Evidence from a field experiment. NBER Working Paper No. 25948. Cambridge, MA: National Bureau of Economic Research.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.20 MB