How can we motivate civil servants to better serve citizens? In this essay, EPoD’s Asim I. Khwaja and Tiffany Simon, with coauthor Adnan Khan, explain EPoD research on civil service reform in Punjab, Pakistan. Experiments with provincial tax departments showed that tax collectors respond well to two types of performance incentives, and this raised revenues – but, they observe, all incentives are not created equal. This piece is included in a conference volume, Pakistan’s Institutions: We Know They Matter, But How Can They Work Better? edited by Michael Kugelman and Ishrat Husain and published by the Wilson Center.     

According to a September 2016 UNDP report, “Pakistani civil services have perhaps been the most often reformed with least effect than anywhere else in the world.” From a historical perspective, a majority of previous civil service reforms in Pakistan over the past 40 years targeted organizational restructuring within the civil service, focusing on adjustment of pay scales; the creation, merger, or disbanding of occupational groups; and changes in functions and powers. While previous reforms have mentioned performance management, they made little progress toward measuring performance in various dimensions or toward new ways of rewarding it. This essay attempts to address this gap and question of effective civil service reforms in Pakistan through a discussion of key principles to consider when designing incentives for civil servants, drawing on our research on tax collection in Punjab, Pakistan.

Conceptually, the principal-agent model is the main analytical framework to address incentive problems that arise when a principal (government) delegates tasks to an agent (bureaucrat) that the principal cannot observe directly. This information gap makes it hard to evaluate staff performance, identify the sources of inefficiency, and design rewards or punishments that benefit the principal. What further complicates designing performance incentives for bureaucrats in countries like Pakistan are the particular constraints on recruitment, monitoring, pay, and promotion faced by their governments.

The key therefore behind effective performance management is to design systems in such a way that they allow one to address the underlying principal-agent problems. In this regard, it is helpful to note that there are two areas in which civil servants need to be incentivized to improve performance and service outcomes.

The first one—and the one most often recognized—is to incentivize the bureaucrat to exert the desired level of effort. Particularly in the developing country context, civil servants are paid modest wages and offered minimal to no performance bonus, providing little motivation for public employees to exert effort in the satisfactory performance of their duties. For example, in the education sector, absent effort (or the physical presence) on the part of the teacher, students will not learn. In this case, an effective incentive would be designed to motivate the civil servant to exert the effort needed to teach students (or in the case of teacher absenteeism, make the effort to attend class).

However, a second equally important—but often overlooked—aspect of performance management systems is creating the right incentives for the civil servant to (truthfully) disclose information to the state. In cases where the civil servant has differential access to information, they may have counterincentives—such as the collection of rents (bribes)—that motivate them to not reveal this information to the state.

For example, police officers are expected to perform the service of enforcing the laws of the state, and to inform the state and take appropriate action when laws are broken. However, in cases where a police officer observes a citizen breaking the law, they may collude with (or extort) the citizen and accept a bribe in return for not revealing this information to the state.
While the importance of effort and truthful information disclosure likely varies across the types of service one considers, both these factors are important enough that any performance management system should consider how to address them. As previously mentioned, the dimension of the primary challenge in education, when considering the teacher-student relationship, is one of effort—that is, the teacher may not put in enough effort to enhance student learning. However, when viewed holistically, we can find an aspect of corruption as well—not between the student and teacher, but between the teacher and the school inspector. While the school inspector should be most concerned with the learning outcomes of students, at the end of the day, they may be better off not revealing the school’s poor performance to the state and accepting a bribe from the teacher instead.

In this essay, we will therefore consider the design of performance-based systems that consider both the effort and information revelation dimensions. We examine specifically the case of the property tax collector in Pakistan, as both dimensions are particularly salient in their situation. Property tax collectors in Pakistan are responsible for assessing the value of property tax to be paid by taxpayers, with little external oversight. Property tax collectors must exert effort to accurately assess tax liabilities and interact with taxpayers. However, they also face an incentive to withhold information to the state—they may receive a side payment from the property owner by colluding with them and incorrectly assessing the true tax liability of the property.
In considering this example, we will derive implications and insights from two related successful multi-year studies that we carried out in collaboration with the Excise and Taxation Department of the government of Punjab. These studies are distinct in that they were a novel collaboration between leading researchers (from Harvard, London School of Economics, and MIT) working through the Center for Economic Research in Pakistan (CERP), and they were also conceptualized, designed, and implemented with the support of a range of government departments—primarily the Punjab Excise and Taxation Department, with support from the Finance, Planning and Development Departments, as well as the Office of the Chief Minister of Punjab.

The first study, already published, focused on using financial incentives to reward performance. The second study uses non-financial rewards—specifically merit-bases postings—as a way of incentivizing civil servants. While this study is not yet published, initial results are extremely promising. Both studies are also the first of their kind. The first study is unique in examining performance pay in taxation using the gold standard of evaluation designs: a randomized controlled trial (RCT). This means that the results represent a rigorous causal impact of the reforms. The second one, to our knowledge, is the first-ever RCT examining the impact of using merit-based postings as an incentive for public sector workers. While we will primarily draw on insights from the first study, we will also utilize the second when relevant.

Context: Property tax in Pakistan     

Although Pakistan’s experience with property taxation is long compared to many other developing states, the yield of property tax in Punjab, Pakistan’s most populous province, is approximately one-fifth the level of collection in comparable countries. A narrow tax base, low tax rates, weak administration, and distrust of public institutions contribute to this problem. Additionally, as in many other developing countries, tax officials play the key role of assessing, auditing, and enforcing taxes. This role, coupled with low salaries and few or no performance incentives, may lead to tax officials accepting payments from property owners for keeping properties off tax records, assessing properties at a lower rate, or permitting erroneous exemptions, reducing tax revenues collected for the state.

Urban property tax rates in Punjab are determined by property size and a formula that takes into account property location, classification—residential, commercial, or industrial—and whether the property is rented out or occupied by its owner. Some exemptions are granted for specific groups, including retired government employees, charitable religious institutions, widows, and the disabled.

Tax collection units are made up of 2,000 to 10,000 properties in geographical areas defined as “circles.” Each circle is assigned a team of three tax officials who are responsible for determining and recording the tax liability of each property and sending an annual tax bill to the property owner. These tax officials are provincial career bureaucrats, who, similar to their counterparts in other developing countries, are paid relatively low wages within a given salary band, with minimal opportunity for performance pay.

Additionally, there are few audit mechanisms and minimal monitoring of the activities of the tax collection staff. The property database is maintained in a physical register, new properties that are registered are not automatically added (nor are building permits or rental agreements), and tax bills are still handwritten and delivered in person by tax staff in some areas, providing multiple occasions for error and manipulation.

Given that tax staff have discretion in determining property assessments and exemptions, tax evasion often takes place through assessments and exemptions that are difficult to verify but may have a large impact on the tax bill. For example, rented residential properties are taxed 10 times more than owner-occupied properties, and commercial properties are taxed three to six times more than residential properties. Both types of assessments are difficult to confirm—one would not be immediately able to assess by sight whether a building is occupied by a renter or an owner, or what percentage of a mixed-use property is used for commercial purposes rather than residential.
 

Good principles of incentive design   
 

Below we highlight key lessons in designing performance incentives for civil servants based on our findings from a series of policy interventions we conducted in collaboration with the Punjab Excise and Taxation Department. Both experiments lasted two years. We focus primarily on an experiment that introduced performance pay for tax collectors in Punjab. While the primary objective of the civil service reform was to increase tax collections, this had to be done taking into account both considerations of fairness (both under- and over-taxation/extortion were to be minimized) and customer satisfaction (taxpayers had to feel that they were treated reasonably). This meant that in addition to tax collection, we judged the success of the reforms in terms of their impact on accuracy of tax assessment and taxpayer satisfaction.

While the findings outlined below are drawn from the specific study we carried out, we believe that the principles we outline below have broader applications—not just for other types of taxes, but for other services as well.
 

Principle 1: Use Performance Incentives     
The clearest lesson from our research on tax collection in Punjab, Pakistan, is that performance incentives work. While this may sound unsurprising, there is nevertheless a substantial amount of debate as to whether incentives can work in a situation where there may be substantial rents. In other words, the concern is that if (corrupt) civil servants are making multiples of their official salary in the form of bribes, then even very attractive performance reward systems may simply not be financially attractive enough to induce a change in behavior. In the Pakistan context, previous wage increase reforms, while not rigorously studied, had been viewed as largely ineffective.

In our experiment, we randomly assigned tax collectors in their property tax circles to one of three performance-pay schemes:

• In the “revenue” scheme, tax collectors were paid a bonus based on the amount of revenue they collected above a three-year average of tax collections in that circle.
• The “revenue plus” scheme mirrored the “revenue” scheme, but adjusted the bonus based on additional information from taxpayer satisfaction surveys and the accuracy of tax assessments.
• The “flexible bonus” scheme awarded bonuses based on criteria set by a Performance Evaluation Committee (PEC) made up of senior tax officials, and allowed for subjective adjustments at the end of the period.

Additionally, there are few audit mechanisms and minimal monitoring of the activities of the tax collection staff. The property database is maintained in a physical register, new properties that are registered are not automatically added (nor are building permits or rental agreements), and tax bills are still handwritten and delivered in person by tax staff in some areas, providing multiple occasions for error and manipulation.

Given that tax staff have discretion in determining property assessments and exemptions, tax evasion often takes place through assessments and exemptions that are difficult to verify but may have a large impact on the tax bill. For example, rented residential properties are taxed 10 times more than owner-occupied properties, and commercial properties are taxed three to six times more than resident
 
We found that over two years, tax revenues in circles where tax collectors were assigned to performance pay schemes had a 46 percent higher rate of growth compared to control groups.
This growth in revenues can be attributed for the most part to an expansion in the tax base. This came from both an increase in the number of taxed properties and the reassessment of properties already on the tax roll. These were accurately recorded and reported.

We also found that despite successfully increasing revenues, the performance schemes had little adverse effect on taxpayer satisfaction. Neither did the taxpayers’ perceptions of the quality of service of the tax office nor their satisfaction with their interactions and dealings with tax officials decline. However, in spite of this rosy picture, we also find that high-powered incentives increase the bargaining power of tax collectors at the cost of taxpayers, which can have potentially undesirable consequences that we discuss further at the end of the essay.

Principle 2: Keep Incentives Simple            

Based on our findings, easy to understand, transparent, ex ante, and objective incentive schemes are the most effective.

In our experiment, we found that the straightforward “revenue” scheme, where tax collectors were paid a bonus directly tied to the revenue they collected above predefined benchmarks, had a 62 percent higher growth rate in total revenue relative to the comparison group by the second year of the experiment. Adding dimensions to performance pay schemes weakened the impact on tax revenue without improving nonrevenue outcomes, such as taxpayer satisfaction.
For example, the “revenue plus” scheme was similar to the revenue scheme, but adjusted the bonus based on additional information from taxpayer satisfaction surveys and the accuracy of tax assessments, with the additional information basically functioning as an audit. While taxpayers expressed slightly greater satisfaction with tax collection under this scheme compared to the other two schemes, “revenue plus” collected substantially less revenue than the “revenue” scheme, with a 41 percent higher growth rate in total revenue relative to the comparison group.

The “flexible bonus” scheme was similar to the “revenue plus” scheme—staff were again divided into three performance groups—however, rankings in these groups were determined by a criteria set by a Performance Evaluation Committee (PEC) made up of senior tax officials. The scheme attempted to mimic bonuses in the private sector, where managers award bonuses from a fixed pool of resources based on factors that may be subjective. This scheme saw no statistically significant increase in current year tax revenue collected, and in fact lost money for the government when comparing the cost of bonus payout to additional tax collected. An explanation for this could be that tax officials in this scheme found the subjective aspects of this incentive scheme not sufficiently motivating, or perhaps they did not find the scheme credible.
 
Principle 3: Be Creative About Incentives—Both Financial and Non-Financial Rewards Matter    
 
While money is a powerful incentive, our second study shows that non-financial incentives can be just as powerful. Transfers of civil servants to new postings (for example, to a different office or different city) occur often in bureaucracies, but are usually attributed to bureaucratic arbitrariness, personal connections, or even preferences of managers. In our second study, we examined whether the ability to transfer to more desirable locations could be offered as a performance incentive to tax officials in Punjab. We randomly assigned tax circles to groups of approximately 10, and informed inspectors in these groups that they would be reassigned to a new circle within that group at the end of the year, based on the improvement of tax revenue within their circle and their rankings of desired postings. While this study is as yet unpublished, our initial findings show that this performance incentive substantially increased tax revenues in these groups compared to comparison groups. This suggests that when designing incentive schemes, one need not always resort to financial incentives—there are often substantial non-pay job attributes, like where one works, that can be very powerful motivators. To the extent that these non-pecuniary reward systems do not require an explicit financial outlay on the part of the government, such schemes may be even more attractive than standard financial reward schemes.
 
Principle 4: Consider the Frequency of Incentives    
 
When designing an incentive strategy, the value of information and the type of effort the civil servant must exert may determine the frequency of your incentive offer.

Although our performance pay experiment found that providing incentives each year for two years was cost-effective (the total cost of paying performance bonuses was less than the total additional revenue collected through the scheme, except in the case of the “flexible bonus”) with a large return on investment for the government, it is not clear that offering performance bonuses on an annual basis is cost-effective in the long run.          

In the case of taxation, a primary reason for low tax collection is due to collusion, with the taxpayer and the tax collector colluding to conceal the true value of the property. Increases in tax revenue following the performance pay incentive in our study are based on the disclosure of this information to the government. Once this information has been revealed to the state, and the property’s true liability has entered the tax base, this property can be accurately taxed for several years before it needs to be reassessed. Therefore, it may not be necessary to pay a bonus annually—instead, bonuses could be awarded every few years. On the other hand, if the increases in tax revenue can be attributed to the addition of new properties to the tax roll (as opposed to accurate reassessments of real tax liabilities of known properties), then an annual bonus might be cost-effective in the long run. Our study found that the increase in tax revenue collected lasted beyond the period of performance pay, suggesting that the former effect is quite important, and that it is more financially beneficial for the government to introduce this scheme every few years.

Turning to our study on merit-based postings, while offering this incentive is practically free for the government, our initial results show that tax collectors who experienced the scheme twice (that is, they were subject to a potential reposting two years in a row based on their performance) were less effective than those tax collectors who only experienced it once. This may be because tax collectors who are transferred find it hard to perform as well in their new circle—they may be less familiar with the circle and not know which properties can be added to the tax rolls.
Alternatively, it could be discouragement—tax collectors who worked hard in their first year in order to be transferred to their circle of choice might find it frustrating to need to exert the same amount of effort in the second year just to avoid losing their new posting. Similar to the performance pay experiment, governments might find the transfer incentive more effective if offered not annually but every few years.

The key message here is that performance pay schemes may not need to be offered each year; in certain cases, it may be more desirable from the state’s perspective to offer such reward schemes every few years. This is determined by whether the benefits generated by providing incentives last only the duration of the incentive period, or whether the benefits tend to last for a while once obtained. Recall that we had mentioned the two salient dimensions for reform were effort and information revelation. When considering the dimension of effort, the incentivized activity needs to be continuously provided to reap the desired outcome (that is, teachers need to work hard every day and show up to class to ensure their students learn). In this case, annual (or even more frequent) performance pay may provide more effective outcomes. However, in cases where the more salient dimension is information—as in our property tax study—the benefits remain once information has been revealed. In this case, performance pay need not be regular, but rather timed in such a way as to correspond to key moments where information revelation is particularly valuable. So in the case of property tax, if it takes X years for a substantial change in property valuations in an area, then it makes sense to introduce a substantial “performance year” every X years.

A few notes of caution        

Rewarding civil servants strengthens their position. While this is clearly desirable—in that it makes it more attractive for them to perform their duties—in a world where rents and collusions may be prevalent, such an effect could have undesirable consequences. Consider the case of property tax again—performance pay increases a tax collector’s motivation to accurately assess and collect taxes and increases their formal salary. However, this also implies that if the status quo is a collusive arrangement, where the tax collector accepts side-payments to lower a taxpayer’s liability, the tax collector now has greater “bargaining power.” Specifically, that tax collector’s position is strengthened vis-à-vis the taxpayer because the higher performance pay from the state means that the tax collector has less to lose if the side-payment negotiations fail. 

This is precisely the feature that results in a breakdown of collusive arrangements with some taxpayers (and hence an increase in taxes collected), and a continuation of collusive arrangements with others, where the going “bribe rate” has gone up. Thus, taxpayers who remain in a collusive arrangement will actually end up paying higher side payments after the civil service reform.

Strengthening the position of a civil servant in a situation where collusion is rampant therefore has the potential for perverse incentives. In the tax example, offering performance pay to tax collectors may incentivize them to extort even more from taxpayers. Indeed we found in our experiment that while on the whole tax collection did increase, this was not because everyone paid more taxes. While some people did pay substantially more taxes, a large fraction of taxpayers did not pay any more taxes but instead reported making higher and more frequent side payments.

Keeping in mind the potential for perverse incentives, it is therefore important to include separate checks against bribery and extortion in the design of any performance incentive. Governments can potentially mitigate undesirable effects from instituting performance pay and the ensuing increase in a civil servant’s bargaining power by developing audit processes or third party data verification systems, particularly supported by the use of information technology.
However, automation and the removal of human interface should not be seen as a panacea. In the case of property tax, digitization and automation are sensible reforms, which can support the state in ensuring truthful reporting by tax collectors. But these processes will not be able to replace the role of the tax collector in recording new properties or changes in the use of properties, for example, from residential to commercial or owner-occupied to rented. These kinds of reforms should therefore be seen as complementary to performance incentives.

In addition, it is key to offer incentives for outcomes that are ultimately the desired objective of the state and cannot be easily manipulated. Returning to the education example, measuring student test scores as performance incentive runs a risk, as test scores can be manipulated. 

Additionally, incentivizing teacher performance based on student test scores creates the possibility of test scores becoming the only indicator educators and other actors care about, potentially glossing over actual student learning and wellbeing. In the case of our property tax study, this meant that we ensured that there were adequate and objective measures of not only taxes collected but also, as previously noted, the desired outcomes of assessment accuracy and taxpayer satisfaction.

Finally, a challenge in any performance scheme is that it is difficult to care about multiple outcomes; to the extent that you desire multiple outcomes, you may need to reward or incentivize more than one outcome. However, as we saw in the variants of the schemes in the first study, adding additional dimensions by which the civil servant’s performance will be judged can weaken the effectiveness of the scheme by making it more complicated and less transparent. Any performance management system will therefore face a tradeoff between making schemes align with the multi-dimensional objectives of the state and the simplicity and credibility of the incentive system.
 
Effective public service delivery ultimately relies on the performance of bureaucracies. Improving bureaucratic performance is all the more important for Pakistan, given the widespread consensus that its bureaucracy is underperforming compared to global standards of modern governance. But in order for civil service reforms to succeed, particularly given the fate of previous attempts, we must design incentive structures that effectively address the underlying principal-agent problem faced by bureaucracies. Our results point to the idea that carefully designed performance incentives can be a politically and financially practical approach to reforming the civil service in Pakistan.  g

Adnan Q. Khan is research and policy director of the International Growth Centre at the London School of Economics and Political Science. Asim I. Khwaja is the Sumitomo-FASID Professor of International Finance and Development and co-director of Evidence for Policy Design at the Harvard Kennedy School. Tiffany M. Simon was a former senior research manager at Evidence for Policy Design.

This essay cites findings from projects that are the result of collaborations among many individuals. The authors thank all the secretaries, director generals, directors, and project directors from the Punjab Department of Excise and Taxation; the Punjab Finance, Planning, and Development Departments; and the Offices of the Chief Secretary and Chief Minister for their support over the multiple years of these projects. They also thank RAs in Cambridge and Lahore for providing outstanding research assistance. The views expressed here are those of the authors, and do not necessarily reflect those of the individuals or organizations acknowledged.