“Qualitative Comparative Analysis” as a tool for modeling agents

“Qualitative Comparative Analysis” (QCA; Rihoux and Ragin, 2009) is a technique used preferably in the field of social science and of political science. It is recommended for the analysis of data sets with a limited number of cases. The mathematical background of the technique is binary logic and Boolean algebra. We test the technique for modeling human decision behavior within an agent-based model (ABM) using R-software and the QCA-package of Thiem and Duşa (2013).

The ABM has been constructed for better understanding the outdoor recreation behavior of inhabitants of Langenthal, a small city located in the Plateau of Switzerland. We studied the effect of different land-cover change (LCC) scenarios on agent’s behavior, and we evaluate options for mitigating negative effects of LCC. The study is based on a survey with 233 persons that was carried out in 2010. Within this study, the interviewees have been asked to mark their preferred recreation areas on a topographic map, showing the surrounding of the city as a grid of 176 square kilometer cells (Degenhardt et al., 2011). In average 13 cells have been marked. This landscape has been analyzed by Kienast et al. (2012). For our analysis, we use seven of the land-cover descriptors they have identified as predictive for recreation preference.

We modeled the behavior of all 233 agents individually in six steps: (1.) The cells marked by an agent have been supplemented by the same number of randomly selected adjacent cells. The land-cover descriptors of all these cells form the input to the model (2.) Descriptor values greater than the median were assigned  value “1”, the others were assigned “0”. (3.) The resulting table, where each cell is described  by seven 0/1-values, has been simplified to a rule (denoted as “minimal formula” in QCA-terminology) that describes agent’s selection. (4.) Then values of land-cover descriptors have been changed according to the LCC scenario and have been re-calibrated using the medians of step 2. (5.) Then for each cell the binary difference to the minimal formula has been computed. (6.) In the last step, agent’s reaction on LCC was simulated: The agent replaces some of the marked cells by adjacent cells with the goal that totalized differences becomes as small as possible.

In our scenario of LCC, in average half of cells marked by an agent became less suitable for recreation. In average a quarter of the cells are subsequently exchanged by the agents. The agents are able to compensate with these exchanges roughly half of the loss of recreation suitability.

We learned that QCA is suitable for modeling agents if knowledge about agents is limited and agents’ decisions are rule-based. QCA also offers options (e.g. “logical remainders” and methods for “resolving contradictory configurations”) for implementing learning behavior and for modifying agents’ decision making.