We proposed a novel approach to investigate how problem-solving strategies, identified using response time and eye-tracking data, can impact individuals’ performance on the Object Assembly (OA) task. To conduct an integrated assessment of spatial reasoning ability and problem-solving strategy, we applied the Multimodal Joint-Hierarchical Cognitive Diagnosis Model (MJ-DINA) to analyze the performance of young students (aged 6 to 14) on 17 OA items. The MJ-DINA model consists of three sub-models: a Deterministic Inputs, Noisy “and” Gate (DINA) model for estimating spatial ability, a lognormal RT model for response time, and a Bayesian Negative Binomial (BNF) model for fixation counts. In the DINA model, we estimated five spatial cognitive attributes aligned with problem-solving processes: encoding, falsification, mental rotation, mental displacement, and intractability recognition. Our model fits the data adequately, with Gelman–Rubin convergence statistics near 1.00 and posterior predictive p-values between 0.05 and 0.95 for the DINA, Log RT, and BNF sub-models, indicating reliable parameter estimation. Our findings indicate that individuals with faster processing speeds and fewer fixation counts, which we label Reflective-Scanner, outperformed the other three identified problem-solving strategy groups. Specifically, sufficient eye movement was a key factor contributing to better performance on spatial reasoning tasks. Additionally, the most effective method for improving individuals’ spatial task performance was training them to master the falsification attribute. This research offers valuable implications for developing tailored teaching methods to improve individuals’ spatial ability, depending on various problem-solving strategies.

Spatial reasoning ability plays a critical role in predicting academic outcomes, particularly in STEM (science, technology, engineering, and mathematics) education. According to the Cattell–Horn–Carroll (CHC) theory of human intelligence, spatial reasoning is a general ability including various specific attributes. However, most spatial assessments focus on testing one specific spatial attribute or a limited set (e.g., visualization, rotation, etc.), rather than general spatial ability. To address this limitation, we created a mixed spatial test that includes mental rotation, object assembly, and isometric perception subtests to evaluate both general spatial ability and specific attributes. To understand the complex relationship between general spatial ability and mastery of specific attributes, we used a higher-order linear logistic model (HO-LLM), which is designed to simultaneously estimate high-order ability and sub-attributes. Additionally, this study compares four spatial ability classification frameworks using each to construct Q-matrices that define the relationships between test items and spatial reasoning attributes within the HO-LLM framework. Our findings indicate that HO-LLMs improve model fit and show distinct patterns of attribute mastery, highlighting which spatial attributes contribute most to general spatial ability. The results suggest that higher-order LLMs can offer a deeper and more interpretable assessment of spatial ability and support tailored training by identifying areas of strength and weakness in individual learners.

Item response theory (IRT) models usually assume that latent traits follow normal distributions. Nonetheless, violations of the normality assumption are not uncommon in psychology and psychiatry. While several studies have investigated the impact of non-normality on parameter estimation for some IRT models, bifactor models for polytomous data have been largely excluded despite their increasing popularity in psychology and education. Existing findings on other IRT models may not apply to bifactor models because of their unique characteristics. In this study, we examined the impact of latent trait non-normality on parameter estimation in the bifactor graded response model using a simulation study. The results indicate that (1) general-factor non-normality primarily affected item parameter estimates associated with the general factor but not specific factors, (2) specific-factor non-normality showed smaller and more localized effects, primarily on bias in specific-factor discrimination and intercept estimates; and (3) despite the misspecification of latent trait priors, maximum a posteriori estimates of latent traits tend to be more stable than the maximum likelihood estimates. Implications of the findings are discussed.

The Rasch Tree model uses model-based recursive partitioning to detect differential item functioning (DIF) across combinations of covariates. This intersectional approach is a powerful exploratory tool, but it has limitations when incorporating unordered polytomous covariates (e.g. race), particularly surrounding interpretability and computational efficiency. As a result, many studies restrict Rasch Tree analyses to contain only dichotomous or continuous covariates. Recent work has proposed collapsing polytomous covariates into fewer categories prior to model fitting, but the effects of this approach have not been assessed. The present study evaluates the impact of collapsing polytomous covariates on Rasch Tree performance using both simulations and empirical data. We compare DIF detection accuracy and model complexity under the default partitioning algorithm and an alternative approach based on k -means clustering. Results indicate that clustering reduces sensitivity and overall DIF detection accuracy. In addition, detection accuracy for polytomous covariates remains low even under the default procedure, highlighting the need for improved methods for incorporating such covariates in Rasch Tree models.