measr

Diagnostic classification models (DCMs) are a class of psychometric models that estimate respondent abilities as a profile of proficiency on a pre-defined set of skills, or attributes. Despite the utility of DCMs for providing fine-grained and actionable feedback with shorter assessments, they have are not widely used in applied settings, in part due to a lack of user-friendly software. Using R and Stan, measr (said: “measure”) simplifies the process of estimating and evaluating DCMs. Users can specify different DCM subtypes, define prior distributions, and estimate the model using the rstan or cmdstanr interface to Stan. You can then easily examine model parameters, calculate model fit metrics, compare competing models, and evaluate the reliability of the attributes.

Installation

You can install the released version of measr from CRAN with:

install.packages("measr")

To install the development version of measr from GitHub use:

# install.packages("remotes")
remotes::install_github("r-dcm/measr")

Because measr is based on Stan, a C++ compiler is required. For Windows, the Rtools program comes with a C++ compiler. On Mac, it’s recommended that you install Xcode. For additional instructions and help setting up the compilers, see the RStan installation help page.

Usage

We can define a DCM using dcm_specify(). This function requires a Q-matrix defining which attributes are measured by each item. We also identify any item identifier columns. Other arguments can be specified to customize the type of model to estimate (e.g., type of measurement or structural model; see ?dcmstan::dcm_specify()). We can then estimate our specified DCM using dcm_estimate(). We supply our specification and our data set, along with any respondent identifiers. As with dcm_specify(), other arguments can be specified to customize the model estimation process (e.g., estimation backend and method; see ?dcm_estimate()).

To demonstrate measr’s functionality, example data sets are available in the dcmdata package. Here we use the Examination of Certificate of Proficiency in English (ECPE; Templin & Hoffman, 2013) data (see ?dcmdata::ecpe for details). Note that by default, measr uses a full Markov chain Monte Carlo (MCMC) estimation with Stan, which can be time and computationally intensive. For a quicker estimation, we could use Stan’s optimizer instead of MCMC by adding method = "optim" to the function call. However, please note that some functionality will be lost when using the optimizer (e.g., the calculation of some model fit criteria requires the use of MCMC).

library(measr)

model_spec <- dcm_specify(dcmdata::dtmr_qmatrix, identifier = "item")

model <- dcm_estimate(
  dcm_spec = model_spec,
  data = dcmdata::dtmr_data,
  identifier = "id",
  seed = 69385,
  refresh = 0
)

Once a model has been estimated, model parameters, respondent classifications, and results of the model fit analyses can then be extracted using measr_extract().

measr_extract(model, "m2")
#> # A tibble: 1 × 3
#>      m2    df  pval
#>   <dbl> <int> <dbl>
#> 1  261.   293 0.909

measr_extract(model, "classification_reliability")
#> # A tibble: 4 × 3
#>   attribute                 accuracy consistency
#>   <chr>                        <dbl>       <dbl>
#> 1 referent_units               0.928       0.878
#> 2 partitioning_iterating       0.924       0.875
#> 3 appropriateness              0.894       0.817
#> 4 multiplicative_comparison    0.924       0.867

Contributions are welcome. To ensure a smooth process, please review the Contributing Guide. Please note that the measr project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.