Really powerful! How ants help to optimise admission tests

Psychological tests are very popular and play an important role in education – for example, in the selection of students. But where do ants come into play here? Of course, we want a maximum amount of information about candidates using the least amount of resources (i.e. testing time). One benefit of shortening tests is an improved candidate experience – and this is where ants can help! So-called metaheuristics or automatic optimisation algorithms offer flexibility in the selection of items or questions.

A prominent example is the Ant Colony Optimisation (ACO) algorithm inspired by the observation of ants. Experiments with ants have shown that after an initial period of wandering, ants find the shortest path to a food source. This is due to a chemical trail of pheromones that each ant leaves behind on its way. Since more ants travel on the shortest route per unit of time than on longer alternative routes, pheromones accumulate along the shortest route. As a result, higher pheromone levels attract more ants until the majority of ants eventually choose the shortest route.

In the context of item selection, it works as follows: First, different combinations of items (= ants) are randomly drawn from the existing pool. A statistical model is then estimated for each set. On this basis, the set is evaluated with respect to a previously determined optimisation function (= shortest route). Different criteria (e.g. internal structure, measurement accuracy, correlations with external criteria) can be included in this optimisation function. The items or questions in the compilation that best meet the optimisation criteria are more likely to be selected in the next iteration. As the number of iterations increases, an increasingly clear pattern emerges. The result is then an efficient selection of items in terms of the aspects to be optimised. The ACO algorithm can also provide valuable guidance on the optimal test length or number of items.

The advantages of using the ACO algorithm go along with two limitations: First, the ACO algorithm finds an efficient, but not necessarily the best version of a test. However, calculating all possible short versions is usually impossible, especially with a large existing pool of items. The ACO algorithm offers a way out of this computational problem. However, even an ACO algorithm is still computationally complex – computations can take hours or even days.

The ACO algorithm is a powerful tool for optimising tests. Its great strength lies in the versatility of the criteria that can be used to select items or questions. This opens up exciting new possibilities for the flexible and efficient design of selection procedures.