“Sorting the Good from the mIERda” – Introducing and Testing the performance of the Multidimensional Insufficient Effort Responding Detection Approach (mIERda)

While the scientific community is increasingly recognizing the relevance of addressing insufficient effort responding (IER) in survey data, the reliance of current IER detection approaches on cut-offs limits their implementation and complexifies the identification, prevalence estimation, and measurement of the impact of IER. The present article illustrates, across four simulation studies, an innovative application of a fair-cut forest algorithm by introducing the multidimensional Insufficient Effort Responding detection approach (mIERda). Built on an unsupervised learning approach, the mIERda was designed to streamline the IER screening process, address methodological gaps associated with current detection approaches, and fit a maximum of survey specifications. In the present article, the discriminative potential of various indicators of IER (Study 1a) and classification accuracy of the mIERda (Study 1b) were assessed and compared to the current gold standard approach across three scales and various settings applied using standard thresholds recommended from the literature. The classification accuracy was then assessed in datasets containing real human data (Study 2) contaminated with simulated IER respondents and when multiple scales are screened at once (Study 3). Overall, we found the mIERda to be more accurate than our application of the gold standard in simulated and human attentive data. The mIERda was also found to have greater levels of balanced accuracy in longer scales and when the severity of IER was higher. The conditions impacting the classification performance of the mIERda are also discussed. Overall, results provide evidence of the criterion validity of the mIERda and position it as an accurate IER detection approach.