Online Program

Combining Surveys with Overlapping Latent Structures (302613)

 View Presentation

Keywords: survey, factor analysis

Over time companies can accumulate a roster of routine surveys that, when administered in succession throughout the year, may result in survey fatigue. Consolidating items into a single survey can result in item fatigue and cognitive "satisficing", where the effort exerted on early questions erodes until only the effort necessary to complete the survey is put forth. Further, statisticians are often tasked with correlating items within and across surveys, resulting in hundreds of uncorrected and unstructured comparisons. Alternatively, the latent factors behind surveys can be used to eliminate redundancies and provide conceptual clarity. In one approach to combining two employee surveys, an inter-item polychoric correlation matrix was created from weighted scores. Matrix data was reduced using a principal components analysis, and a parallel analysis was used to generate random data eigenvalues for evaluation of the real data eigenvalues. Selected items were assessed with Cronbach’s alpha and a new abbreviated survey was created with 9 factors. Supervisors received results in the form of 9 factors rather than an overall average or a set of individual item scores.