One of the core challenges of decision research is to identify individuals’ decision strategies without influencing decision behavior by the method used. Bröder and Schiffer (2003) suggested a method to classify decision strategies based on a maximum likelihood estimation, comparing the probability of individuals’ choices given the application of a certain strategy and a constant error rate. Although this method was shown to be unbiased and practically useful, it obviously does not allow differentiating between models that make the same predictions concerning choices but different predictions for the underlying process, which is often the case when comparing complex to simple models or when comparing intuitive and deliberate strategies. An extended method is suggested that additionally includes decision times and confidence judgments in a simultaneous Multiple-Measure Maximum Likelihood estimation. In simulations, it is shown that the method is unbiased and sensitive to differentiate between strategies if the effects on times and confidence are sufficiently large.

One major challenge to behavioral decision research is to identify the cognitive processes underlying judgment and decision making. Glöckner (2009) has argued that, compared to previous methods, process models can be more efficiently tested by simultaneously analyzing choices, decision times, and confidence judgments. The Multiple-Measure Maximum Likelihood (MM-ML) strategy classification method was developed for this purpose and implemented as a ready-to-use routine in STATA, a commercial package for statistical data analysis. In the present article, we describe the implementation of MM-ML in R, a free package for data analysis under the GNU general public license, and we provide a practical guide to application. We also provide MM-ML as an easy-to-use R function. Thus, prior knowledge of R programming is not necessary for those interested in using MM-ML.

We present a classification methodology that jointly assigns to a decision maker a best-fitting decision strategy for a set of choice data as well as a best-fitting stochastic specification of that decision strategy. Our methodology utilizes normalized maximum likelihood as a model selection criterion to compare multiple, possibly non-nested, stochastic specifications of candidate strategies. In addition to single strategy with “error” stochastic specifications, we consider mixture specifications, i.e., strategies comprised of a probability distribution over multiple strategies. In this way, our approach generalizes the classification framework of Bröder and Schiffer (2003a). We apply our methodology to an existing dataset and find that some decision makers are best fit by a single strategy with varying levels of error, while others are best described as using a mixture specification over multiple strategies.

Glöckner and Bröder (2011) have shown that for 77.5% of their participants’ decision making behavior in decisions involving recognition information and explicitly provided additional cues could be better described by weighted-compensatory Parallel Constraint Satisfaction (PCS) Models than by non-compensatory strategies such as recognition heuristic (RH) or Take the Best (TTB). We investigate whether this predominance of PCS models also holds in memory-based decisions in which information retrieval is effortful and cognitively demanding. Decision strategies were analyzed using a maximum-likelihood strategy classification method, taking into account choices, response times and confidence ratings simultaneously. In contrast to the memory-based-RH hypothesis, results show that also in memory-based decisions for 62% of the participants behavior is best explained by a compensatory PCS model. There is, however, a slight increase in participants classified as users of the non-compensatory strategies RH and TTB (32%) compared to the previous study, mirroring other studies suggesting effects of costly retrieval.

One major statistical and methodological challenge in Judgment and Decision Making research is the reliable identification of individual decision strategies by selection of diagnostic tasks, that is, tasks for which predictions of the strategies differ sufficiently. The more strategies are considered, and the larger the number of dependent measures simultaneously taken into account in strategy classification (e.g., choices, decision time, confidence ratings; Glöckner, 2009), the more complex the selection of the most diagnostic tasks becomes. We suggest the Euclidian Diagnostic Task Selection (EDTS) method as a standardized solution for the problem. According to EDTS, experimental tasks are selected that maximize the average difference between strategy predictions for any multidimensional prediction space. In a comprehensive model recovery simulation, we evaluate and quantify the influence of diagnostic task selection on identification rates in strategy classification. Strategy classification with EDTS shows superior performance in comparison to less diagnostic task selection algorithms such as representative sampling. The advantage of EDTS is particularly large if only few dependent measures are considered. We also provide an easy-to-use function in the free software package R that allows generating predictions for the most commonly considered strategies for a specified set of tasks and evaluating the diagnosticity of those tasks via EDTS; thus, to apply EDTS, no prior programming knowledge is necessary.

Research on the processing of recognition information has focused on testing the recognition heuristic (RH). On the aggregate, the noncompensatory use of recognition information postulated by the RH was rejected in several studies, while RH could still account for a considerable proportion of choices. These results can be explained if either a) a part of the subjects used RH or b) nobody used it but its choice predictions were accidentally in line with predictions of the strategy used. In the current study, which exemplifies a new approach to model testing, we determined individuals’ decision strategies based on a maximum-likelihood classification method, taking into account choices, response times and confidence ratings simultaneously. Unlike most previous studies of the RH, our study tested the RH under conditions in which we provided information about cue values of unrecognized objects (which we argue is fairly common and thus of some interest). For 77.5% of the subjects, overall behavior was best explained by a compensatory parallel constraint satisfaction (PCS) strategy. The proportion of subjects using an enhanced RH heuristic (RHe) was negligible (up to 7.5%); 15% of the subjects seemed to use a take the best strategy (TTB). A more-fine grained analysis of the supplemental behavioral parameters conditional on strategy use supports PCS but calls into question process assumptions for apparent users of RH, RHe, and TTB within our experimental context. Our results are consistent with previous literature highlighting the importance of individual strategy classification as compared to aggregated analyses.

Adaptive actors must be able to use probabilities as decision weights. In a computerized multi-attribute task, the authors examined the decisions of children (5–6 years, n = 44; 9–10 y., n = 39) and adults (21–22 y., n = 31) in an environment that fosters the application of a weighted-additive strategy that uses probabilities as weights (WADD: choose option with highest sum of probability-value products). Applying a Bayesian outcome-based strategy classification procedure from adult research, we identified the utilization of WADD and several other strategies (lexicographic, equal weight, naïve Bayes, guessing, and saturated model) on the individual level. As expected based on theory, the prevalence of WADD-users in adults was high. In contrast, no preschoolers could be classified as users of probability-sensitive strategies. Nearly one-third of third-graders used probability-sensitive strategies.