Brenton Kenkel — Research

How do social divisions in a piece of territory affect the value and nature of occupying it? Is a conquering power better off when social groups spend their effort fighting each other instead of resisting conquest, as the traditional logic of divide et impera or "divide and conquer" would suggest? To examine these questions, I set up a formal model of conquest in which a conqueror makes extractive demands, and groups in the occupied territory respond by allocating labor between production, resistance, and competition with the other groups. Contrary to the divide et impera logic, the results suggest that an extractive outside force is better off facing a relatively unified society than one with many group divisions. As the number of groups and conflict among them increases, the conqueror is left with a larger share of a diminished economic surplus in equilibrium, and the overall effect on the conqueror's utility is negative. I also find that occupations can be more extractive and violent when groups have clearly defined comparative advantages than when they are essentially similar.

It is well known that diplomatic communication between adversaries is hampered by incentives to misrepresent, while the problem of diplomacy among allies has received much less attention. Using a model of public good provision with a cheap talk extension, I find that incentives to misrepresent are also pervasive in communication between states with common goals: there is no equilibrium in which a state's announcement of how much it is willing to contribute has any effect on its partner's behavior. I discuss the implications of these findings for the study of crisis diplomacy and consider how the model's assumptions would need to be relaxed for informative communication to be possible.

I provide a general method to obtain logistic regression coefficients when there is nonignorable missingness in the response variable. The method makes no assumptions about the missingness mechanism and requires no application-specific programming. The key is to treat the model parameters as partially identified and estimate bounds on the set of regression coefficients that are consistent with the data, as opposed to the traditional approach of obtaining a single point estimate. Drawing from the partial identification literature in econometrics, I develop three closely related methods for estimating bounds on the parameters of a logistic regression model when the missingness mechanism is wholly unknown to the analyst. These methods are computationally intensive but straightforward to implement. I apply the method to data from a published study on insurgency outcomes and find that the original results are robust to assumptions about the treatment of "draws."

For replication code, see my project page on Github.

I develop a model of electoral competition in which candidates' ability to raise money is related to their private information about the policy they will implement if elected. I use the model to analyze how politicians' fundraising decisions are affected by concerns about signaling their policy intentions, and to query whether campaign finance reform can reduce the possibility of electing well-funded candidates with policy intentions far from the median voter. I find an equilibrium of the model and show that it is unique among equilibria that satisfy the D1 refinement. If centrist candidates can raise money more easily than others, they always exploit this advantage in equilibrium, using high spending as a signal of their proximity to the median voter. The reverse is not true when non-centrists have the fundraising advantage. In this case, candidates who spend highly are perceived as having extreme policy intentions, offsetting the otherwise positive electoral effect of spending. The candidates who face this dilemma resolve it either by imitating the strategy of their centrist peers or by spending enough to compensate for their views being revealed. The electoral consequences of campaign finance reform are also asymmetric. When non-centrists can raise money more easily, a marginal decrease in the size of their advantage may increase the chance that a centrist is elected; however, in the reverse case, such measures have no effect on the chance of electing a non-centrist.

We introduce a new statistical model to estimate the shape of potentially nonlinear, multivariate relationships. The method is built on basis regression, in which simple functions like polynomials are combined to approximate more complex relationships. To reduce the instability typically associated with basis regression, we use penalized regression techniques that perform automatic model selection, eliminating many terms from the final estimate. We focus on methods that satisfy the oracle property, which guarantees that terms with no true effect on the outcome are excluded from the estimated model in sufficiently large samples. Finally, we calculate standard errors and other estimates of variability via the bootstrap. In a series of simulations, we show that our method can accurately estimate nonlinear relationships, even if the exact functional form is not known in advance. We apply our method to Gartzke's (2007) data on the "capitalist peace" and find that joint democracy and trade dependence may increase the chance of conflict in some cases, contrary to the original results.

The standard practice when estimating a treatment effect is to include all available pre-treatment variables in the propensity score, and we demonstrate that this approach is not always optimal when the goal is bias reduction. We characterize the conditions under which including an additional relevant variable in the propensity score increases the bias on the effect of interest across a variety of different implementations of the propensity score methodology. Moreover, we find that balance tests and sensitivity analysis provide limited protection against overadjustment.