A cognitive psychometric model for assessment of picture naming abilities in aphasia

Abstract

Picture naming impairments are a typical feature of stroke-induced aphasia. Overall accuracy and rates of different error types are used to make inferences about the severity and nature of damage to the brain's language network. Currently available assessment tools for picture naming accuracy treat it as a unidimensional measure, while assessment tools for error types treat items homogenously, contrary to findings from psycholinguistic investigations of word production. We created and tested a new cognitive psychometric model for assessment of picture naming responses, using cognitive theory to specify latent processing decisions during the production of a naming attempt, and using item response theory to separate the effects of item difficulty and participant ability on these internal processing decisions. The model enables multidimensional assessment of latent picture naming abilities on a common scale, with a relatively large cohort for normative reference. We present the results of 4 experiments testing our interpretation of the model's parameters, as they apply to picture naming predictions, lexical properties of the items, statistical properties of the lexicon, and participants' scores on other tests. We also created a website for researchers and clinicians to analyze item-level data using our model, providing estimates of latent abilities and percentile scores, as well as credible intervals to help gauge the reliability of the estimated model parameters and identify meaningful changes. To the extent that the model is successful, the estimated parameter values may aid in treatment decisions and progress monitoring, or they may help elucidate the functional properties of brain networks. (PsycINFO Database Record

Figure 1

The MPT model architecture. Nodes with rounded corners represent latent processing decisions, and leaf nodes with square corners represent response types. C = Correct, S = Semantic, F = Formal, M = Mixed, U = Unrelated, N = Neologism, AN = Abstruse Neologism, NA = Non-naming Attempt. Each branch is associated with a probability indicated by the letters a–h.

Figure 2

Frequency distributions of posterior means for the MPT model’s ability and difficulty parameters. Although Sem is a probability, it was converted to a logit scale for consistency with the other parameters.

Figure 3

Frequency distributions of posterior 95% credible interval (CI) widths for the MPT model’s ability and difficulty parameters. Although Sem is a probability, it was converted to a logit scale for consistency with the other parameters.