We appreciate Tsai’s letter, and we acknowledge that, in our recent article, we provided an oversimplification of a complex topic. We did so to present a clear argument to those without formal technical training in statistics. We apologize for any confusion our format may have caused, and we would like to take this opportunity to address Tsai’s points as well as to provide further clarification.
We agree that multivariate statistics is a broad area,1 but the purpose of our article was to encourage authors to use the term “multivariate” correctly and not to use it when the models they describe are really multivariable. The most important point to consider is that we are discussing the nuance in the use of terminology in the context of regression only, and not in the broader context of the type of statistical analysis conducted. We believe that the terms “univariable” and “multivariable” should only be used in the regression context to describe the number of predictors in the model, whereas the terms “univariate” (1 variable), “bivariate” (2 variables), and “multivariate” (multiple variables) should be used to describe the type of statistical analysis being conducted.2 Thus a t-test is a type of bivariate statistical analysis because of the use of two variables.3
We concur that multivariate models can be linear, logistic, or proportional hazards, and we did not mean to suggest that these were mutually exclusive. These terms are simply how we think regression models should be described in the public health literature. For example, if authors use the terms “linear” or “logistic regression,” then “univariate” is implied. But they should also specify whether the model is simple or multivariable. If they use a multivariate regression model, they should still specify the type of model (e.g., linear, logistic, or proportional hazards) and whether it is unadjusted (simple) or adjusted (multivariable). We appreciate Tsai’s affirmation that the terms “multivariate” and “multivariable” should not be used interchangeably as they have two distinct meanings and that some regression models would be most appropriately defined as multivariate multivariable models.
Until we reach a consensus on how to describe regression models, we encourage readers to make sure they understand what is meant when the term “multivariate” is used to define a regression model.
Acknowledgments
B. Hidalgo was supported in part by a postdoctoral training grant from the National Heart, Lung, and Blood Institute (grant T32HL072757). M. Goodman was supported by the Siteman Cancer Center, the National Cancer Institute (grant U54CA153460), and the Washington University School of Medicine Faculty Diversity Scholars Program.
References
- 1.Marcoulides GA, Hershberger SL. Multivariate Statistical Methods: A First Course. Mahwah, NJ: Lawrence Erlbaum Associates; 1997 [Google Scholar]
- 2.Diamantopoulos A, Schlegelmich BB. Taking the Fear Out of Data Analysis: A Step-by-Step Approach. London, UK: Thomson Learning; 1997 [Google Scholar]
- 3.Sims RL. Bivariate Data Analysis: A Practical Guide. Hauppauge, NY: Nova Science Publishers; 2004 [Google Scholar]