Data Censoring
When researchers only have partial information about the values of some data points (knowing that they are at least as large as some value or no larger than some other value), the correlation between those variables is distorted. For example, when exact opposites divide dimension into two parts, each variable is 0 when the other is positive and so they do not correlate -1.
Barchard, K. A., & Russell, J. A. (2021, February 11-13). Correlating positively and negatively keyed items: The problem and the solution. [Poster presentation]. Society for Personality and Social Psychology Virtual Annual Convention.
This poster shows that the R package lava provides precise and accurate estimates of the correlations between uncensored variables, when given the data from censored variables, unless there is severe censoring. Poster
Barchard, K. A., & Russell, J. A. (2020, October 28 – 31). Why factor analyses with negatively keyed items fail. Poster presented at the Western Psychological Association Annual Convention, San Francisco CA.
This poster explains that negatively keyed items fail to load on the same factors as positively keyed items because of censoring and because of ambiguity at the disagreement end of response scales. It explains how to eliminate gradations of disagreement and to use censored data analysis methods. Poster Handout Presentation
Barchard, K. A., & Russell, J. A. (2020, June 1 – September 1). I’m less and less happy until finally I’m sad: Estimating correlations when variables divide a construct into parts. Poster presented at the Association for Psychological Science poster showcase, Chicago, IL. Poster Handout
This poster describes a method of estimating the correlation between the two variables when each of their measures may have some degree of left or right censoring. To implement this method, use one of the following Excel sheets.
Small file, fast download: Barchard, K. A. (2021, January). CensorCorr: Examining the effect of censoring on correlations (n = 10,000). [Excel file]. https://barchard.faculty.unlv.edu/research/examining-opposites/
More accurate results: Barchard, K. A. (2021, January). CensorCorr: Estimating the effect of censoring on correlations (n = 500,000). [Excel file]. https://barchard.faculty.unlv.edu/research/examining-opposites/