Competing for Global Capital Or Local Voters?

Competing for Global Capital or Local Voters?

The Politics of Business Location Incentives

Online Appendix

Online Appendix I: Balance between City-Level Political Institutions

Although treatment and control have very similar levels of per capita income and economic structures, Table A1 shows that mayor-council systems are significantly: 1) more common in populous, suburban cities; 2) less likely to compete for foreign projects; 3) more concentrated in the North Central region of the United States.

More non-balance is observed on economic policy choices. These powerful mayors are more likely to have development plans and development corporations. By contrast, executives in council-manager systems are more likely to invest in environmental sustainability and cultural programs. Because these are post-treatment variables, measuring decisions made by officials after the establishment of electoral institutions, they cannot (and should not) be addressed in our regression analysis. We present them because they illustrate an unmistakable pattern that aligns with our theory. Mayor-council systems are associated with policies favoring short-term incentives for commercial development, whereas executives in council-manager systems are associated with policies that have more diffuse economic benefits. This table presents non-balance on observable, pre-treatment characteristics, however, it indicates that omitted variable bias is a severe threat to statistical inference.

Online Appendix II: between States with and without Default Mayor Clauses

Our identification strategy for mayor-council institutions employs state laws making mayor-council systems the default form of government. An important clarification is in order about our use of the Nelson data. Our measure is merely one indicator from a collection of indicators that Nelson (2010, 2011) uses to construct an index for 49 US states based on the number of variations of form of government by state law. The ultimate purpose is to examine how state laws shape the ability of municipalities to form hybrid governments (Nelson 2011). Although Nelson concludes that hybrid governments are quite rare, she finds that state laws shape the ability of municipalities to shape their local institutions.

This complete index, while appropriate for Nelson’s purpose, is problematic in our study. First, we are not interested in the customization of institutions. Our concern is the endogenous choice of selecting mayor-council institutions over council-manager institutions or vice versa. Second, some of the indicators that she includes, such has “home rule,” not only affect institutions; they also affect the fiscal discretion of political leaders. This clearly violates the exclusion restriction for an instrumental variables analysis. Consequently, we prefer to focus on a single dimension of state laws governing municipal form of government that we can justify for our instrumental variable regression analysis.

In Figure A2, we map the states that had default clauses in 2004 (midway through our sample), while in Table A2, we provide balance tests using census data. In Table A3 we provide descriptive statistics by state.

Figure B1: States with Default Mayor Regulations

(Green Share= States with Default Mayors, Source (Nelson 2011))

A

Table A4: Entropy Balancing
(Treatment =State has Default Mayor Policy)

A

Appendix III: Robustness Tests

Our main results in the paper using Entropy Balancing are robust to alternative methods. We present additional tests in this appendix.

Incentive Offers

In Models 1 and 2 of Table A5, we first estimate the probability that a municipality offered an incentive during the survey year using a probit specification. It is immediately clear that the size of the coefficient for the dummy variable for mayor-council institutions (Elected Mayors)—the key independent variable—is effectively zero. The bivariate relationship is insignificant and the size of the coefficient declines significantly with additional controls (Model 2). Clearly, the propensity to offer an incentive is driven by structural factors such as population size, regional location, and characteristics of the municipality (e.g., a suburb or metro area).[1] This makes sense, as we noted above, because almost all cities, regardless of electoral institutions, now offer some form of incentives. What really matters is how much they offer and the oversight of their programs. City-years offering zero incentives were dropped from the analysis in Models 3–7.

This means that 1,284 of our projects were allocated to the 837 municipalities that did offer incentives. Some of these municipalities show up multiple times in the project data, as they offered incentives to multiple projects in a given year, which is why our n is larger than the number of cities. Of the cities that offered incentives, the median number of incentivized projects was 1 per year and the mean was 1.18 per year, with 404 offering incentives to more than one project. The number of incentives per municipality at the 95th percentile was 3, with one location offering 95 targeted incentives during the calendar year.[2] Structuring the data at the project level allows us to focus on the behavior of the municipalities that did offer an incentive, which allows us to ascertain whether the amount of money allocated for individual projects was higher for mayor-council systems.

<Insert Table A5 About Here>

To this end, in Models 3 to 7, we regress the natural log of the value of the incentive package that was offered by a municipality on Elected Mayors. [3] Model 3 provides the bivariate relationship, controlling only for survey-year fixed effects to make sure that we wipe out any trending in the use of incentives over time. We use ordinary least squares (OLS) with errors clustered at the municipal-level to account for non-independence among projects in the selection of electoral institutions and the use of incentives. We find that mayor-council systems offer an additional 29% in incentives per project than other institutions.

In Model 4, we include a battery of control variables at the level of the municipality and project. At the municipal level, we use controls at the city level discussed above (e.g., population, metro area, suburb, foreign competition, region of the country (e.g., Northeast), etc.). In addition, we include a number of control variables at the project level. First, we include the natural log of the total number of new jobs created (Jobs) and the log of capital expenditures (Capex). We also include a dummy variable if the investment was in the energy or electricity industry (an industry that is granted more generous incentives than other industries) and a dummy variable if this is a new investment (New) as opposed to an expansion project.

Our control variables behave as expected. Larger investments, in terms of jobs and capital expenditures receive larger incentive allocations, along with new investments and those in the energy sector. Our key independent variable, Elected Mayors, remains statistically significant and substantially large. Mayors in mayor-council systems offer an additional 32% in incentives per investment, accounting for the size, sector, and type of investment. The results are robust to the inclusion of sector fixed effects in Model 5 and state fixed effects in Model 6. In Model 7, we include additional controls for the level of unemployment, existing tax policies, and municipal development plan. We run these last because of concerns that these variables may post-date the allocation of incentives.

Given the fact that an incentive was offered, it is clear that these mayors pay far greater amounts for similarly situated projects. In dollar terms, the difference amounts to $172,242 more in incentives on an average project than council-managers.

One concern is that although we are comparing cities of the same population, region, and even level of economic development program professionalism, the underlying size of municipal budgets can largely shape the ability of local leaders to offer incentives. Using survey data on the size of the annual economic development budget, we run models 8 & 9 that scale the size of the incentives as a percentage of the local economic development budget. Although this variable is the most theoretically appropriate measure, it has a large number of missing values and thus we present this primarily as a robustness test of our original estimates. Again, mayor-council institutions are associated with a larger percentage of their economic development budget being offered as incentives.

Incentive Oversight

With the first two dependent variables (Models 1–10 in Table C2), we use a probit specification, coding the dependent variable as 1 for the existence of the oversight program and 0 otherwise. Models 11–15 assess the number of criteria using OLS. All independent variables are measured in the same year of the survey unless otherwise noted.

As expected, there is considerable variation across municipalities in just this minimum requirement. Although 72% of managers in our sample indicate that they always perform a cost-benefit analysis of incentives, only 59% of municipalities have specific performance requirements.[4] Looking directly at the number of performance criteria, 21 cities employ all six items, 939 employ none, and the average city employs about one item.

We present our regressions in Table A6. All models are estimated with robust standard errors clustered at the municipal level and we report marginal probabilities for easier interpretation of the coefficients. Models 1 and 6 test how mayor-council institutions affect performance without any control variables. Even without any control variables, we find that mayor-council systems are associated with an over 10.5% decline in the probability of having performance requirements on incentives (Model 1) or conducting a formal cost-benefit analysis (Model 5). As our H2 predicts, electoral institutions lead to a decrease in the probability of oversight mechanisms. Our results are similar when we include survey-year fixed effects in Models 2 and 6.

<Table A6 about Here>

In Models 3 and 8 we include our control variables. Although few of our control variables are statistically significant, two variables have a large and statistically significant effect on oversight. First, municipalities with larger populations are more likely to have greater oversight of incentives. More interestingly, foreign competition, or at least the perception of competition with foreign localities, leads to an increased use of oversight mechanisms. The substantive effects of performance requirements and cost-benefit analyses decline slightly with the inclusion of these control variables. Municipalities with mayor-council systems are 10.5% less likely to require performance criteria for incentives and 7.2% less likely to require a cost-benefit analysis of incentive programs. Looking at the regressions on the number of criteria, we find that mayor-council systems employ about 0.33 fewer criteria than council-manager systems. These results are robust across the different specifications, including regional fixed effects (Models 4, 9, and 14) and the additional control variables for the unemployment rate and existing tax policies (Models 5, 10, and 15).

These results support H2, although with an important caveat—they are based on a survey of municipalities, which could be prone to a number of potential measurement problems. Self-reported oversight and perceptions of “foreign competition” can suffer from issues of lying or perception biases. These errors, however, are more likely to bias against a significant coefficient on mayor-council systems, either because of attenuation bias or because mayoral systems are likely to exaggerate the level of oversight.

Table A5: Electoral Institutions and the Average Size of Tax Incentives
(Project-Level)

Table A6: Association between Electoral Institutions on Performance Criteria

A

[1] Unfortunately, project-specific controls (e.g., sector, size, etc.) are not available for projects that were not offered incentives, so we are forced to rely only on municipal-level covariates. The marginal effect should therefore be treated as tentative.

[2] We drop this outlier from our analysis, reducing our n to 1,176.

[3] We considered treating no incentives as $0 and estimated the full-model of the total value, but we discarded this option, as we believed it would bias in favor of our hypothesis, because mayoral systems are marginally more likely to offer incentives in the first place.

[4] The most common performance requirement is based on the number of new jobs created.