Thomas Plümper

 

The Politics of Natural Disasters

 

Natural disasters are caused by natural, geological processes. However, that does not mean that political and economic factors do not influence mortality and damage. Quite to the contrary: the impact of political and economic factors on the number of fatalities and the economic damage is large and significant. While hazards are natural, disasters are not.

 

In many cases, natural hazards become disasters because of political neglect. Governments have very few incentives to respond to risks which are not very likely to materialize in the future foreseeable for politicians -- the time to the next elections in democracies or the time to the next coup or revolution in autocracies.

 

Our research demonstrates that disasters are more deadly and more costly when they do not occur on a regular basis, when governments dominantly depend on few influential members of the elite, and when the effects of natural disasters affect the supply of the population with food, water, and medicine.

 

 

Spatial Dependence

 

The decisions, actions, preferences and beliefs of human beings do not only depend on their context situation, but also on decisions, actions, preferences, and beliefs  of others. The project links recent developments on spatial econometric methodology to theories of spatial dependence and explains how to specify empirical models so that they match theoretical predictions. We focus on spatial-y models -- models that explain outcome in one unit of analysis by outcomes in others. Examples from our own applied research include counterterrorist policies, tax policies, military spending, international treaties, and so on.

The causal mechanism identified by the literature is usually some form of voluntary or involuntary learning or externalities. Obviously, military spending of one country affects the perception of safety of other countries either positively or negatively. Appropriate model specification of spatial dependencies needs to focus on the connectivity matrix, which models the causal mechanism. Traditionally, the connectivity matrix has dominantly been neglected and merely modeled as either distance or contiguity. However, geographical information is at best at functional proxy for the causal mechanism underlying spatial dependence. These causal mechanisms are always a form of interaction, exchange, communication, transaction, observation and so on.

 

 

Effect Size Computation in Non-Linear Models

 

The interpretation of substantive effect sizes in ordinary least squares and other linear estimation models is straightforward: the estimated coefficient of a variable is its marginal effect and the coefficient multiplied for any assumed change in x gives the predicted counterfactual effect. The interpretation of estimation results and the derivation of inferences has become much more difficult because the convenient simplicity of linear models does not carry over to non-linear estimation models. In these model, the effect size of a variable is a function not just of the coefficient of the variable but also of the level of the variable of interests and of the effects of all control variables.

The project analysis the options available to researchers computing the effect sizes of non-linear models. Social science methodology disagrees about the definition and the optimal algorithm and procedure for computing effect sizes. Currently, three broad techniques are used in applied research: average marginal effects, adjusted predictions, and predicted counterfactuals. The project analyzes the properties of these techniques and shows that individual predicted counterfactuals offer the most reliable technique for computing effect sizes.

Thomas Plümper 2014-2017