The Tower of London is commonly used to assess planning ability. Deficient outcomes may however have different causes: A participant may not have the ability to think a sufficient number of steps into the future, or may become, for example, impatient to evaluate different possible paths. Outcomes are thus not pure measures of the "planning" construct of primary interest, which may have contributed to findings of low reliability and low validity of these outcomes in the literature. The advent of computerized testing combined with computational modeling potentially allows to go beyond traditional outcomes such as "total number of moves" and "total time taken" and disentangle different processes that are of primary interest. The goal of the current study is to establish whether a model that consists of "planning ability" and "response inhibition" parameters can be used to describe Tower of London data.

We constructed an algorithm that produces Tower of London data, and a computational model that uses every single decision of a participant as input (e.g., whether a participant moves the red or the blue ball to the right peg in setting 15 when trying to get to setting 28). There are 210 unique decision situations that participants can encounter. Our algorithm and Bayesian hierarchical model uses two parameters for each participant as well as a guessing rule, that together determine the participant's decision at every step. The appropriateness of the model was evaluated in a simulation study, where the simulated distribution of data implied by this model is compared to the empirical distribution of total number of moves observed in real datasets. Data were simulated for 10 items with a sample size of 200 participants.

Our simulation study shows that with our model the empirical distribution of total number of moves is successfully replicated in the distribution of the simulated data.

Computational modeling provides a new window into Tower of London performance by identifying different processes. Modeling thus allows us to go beyond aspecific descriptions of planning ability. Furthermore, using the high-resolution data of computerized testing allows us to estimate these parameters reliably without requiring "big data", keeping participant burden low. This study will be followed up in three ways. First, predictions will be preregistered and tested for these new cognitive outcomes in several large oncological patient samples. Second, the model will be extended to include reaction times, to include an additional metric of cognitive computation. Third, the new cognitive process outcomes will be analyzed in conjunction with cognitive process outcomes on other tests to establish process communalities.