The majority of conflict studies now analyze subnational geographic units below the country level to explain and predict conflict dynamics. This development has revolutionized conflict studies over the last decade, facilitated by the advent of spatially and temporally disaggregated event data. In addition to first- or second-order administrative regions, the PRIO-GRID, which divides the globe into temporally stable, equally sized rectangular cells, has been widely used in both explanatory and predictive conflict research as well as policy applications. However, research using PRIO-GRID largely overlooks the potential threat posed by the modifiable areal unit problem (MAUP) to analyses conducted at the grid cell level. Different spatial units of analysis may yield divergent results due to varying aggregations of data, such as conflict events. We address this gap by introducing a flexible procedure for aggregating PRIO-GRID cells to test the sensitivity, reliability, and validity of statistical results. Our main theoretical contribution is the expectation that bias resulting from the MAUP differs across conflict types. We argue that the spatio-temporal link between the causes of conflict and the geographic manifestation of (non-)violence varies across conflict types. Specifically, we find a stronger spatio-temporal link for secessionist or herder conflicts, whereas this link is weaker for protests and conflicts over government control, increasing the bias resulting from the MAUP. We reproduce several published works to analyze the robustness of previous findings across conflict types. Our findings contribute to the literature on disaggregated conflict studies at the grid cell level.