Abstract
Regression analyses are perhaps the most widely used statistical tools in medical research. Centring in regression analyses seldom appears to be covered in training and is not commonly reported in research papers. Centring is the process of selecting a reference value for each predictor and coding the data based on that reference value so that each regression coefficient that is estimated and tested is relevant to the research question. Using non‐centred data in regression analysis, which refers to the common practice of entering predictors in their original score format, often leads to inconsistent and misleading results. There is very little cost to unnecessary centring, but the costs of not centring when it is necessary can be major. Thus, it would be better always to centre in regression analyses. We propose a simple default centring strategy: (1) code all binary independent variables +1/2; (2) code all ordinal independent variables as deviations from their median; (3) code all ‘dummy variables’ for categorical independent variables having m possible responses as 1−1/m and −1/m instead of 1 and 0; (4) compute interaction terms from centred predictors. Using this default strategy when there is no compelling evidence to centre protects against most errors in statistical inference and its routine use sensitizes users to centring issues. Copyright © 2004 Whurr Publishers Ltd.
Keywords: regression, centring, multicollinearity
Full Text
The Full Text of this article is available as a PDF (181.5 KB).
References
- Aiken LS, West SG. Multiple Regression: Testing and Interpreting Interactions. Newbury Park CA: Sage Publications, 1991. [Google Scholar]
- Appelbaum MI, Cramer EM. Some problems in the nonorthogonal analysis of variance. Psychological Bulletin. 1974; 81: 335–43. [Google Scholar]
- Cohen J. Partialed products are interactions; partialed powers are curve components. Psychological Bulletin 1978; 85: 858–66. [Google Scholar]
- Cohen J, Cohen P, West S, Aiken L. Applied Multiple Regression/correlation Analysis for the Behavioral Sciences. Hillsdale NJ: Lawrence Erlbaum Associates, 2003. [Google Scholar]
- Cramer EM, Appelbaum MI. Nonorthogonal analysis of variance‐once again. Psychological Bulletin 1980; 87: 51–7. [Google Scholar]
- Flack VF, Chang PC. Frequency of selecting noise variables in subset regression analysis: a simulation study. The American Statistician 1987; 41: 84–6. [Google Scholar]
- Glantz SA, Slinker BK. Primer of Applied Regression and Analysis of Variance. New York: McGraw‐Hill, 2001. [Google Scholar]
- IHDP . Infant Health and Development Progam: enhancing the outcomes of low birth weight, premature infants: a multisite randomized trial. Journal of the American Medical Association 1990; 263: 3035–42. [DOI] [PubMed] [Google Scholar]
- Kraemer HC, Stice E, Kazdin A, Kupfer D. How do risk factors work together to produce an outcome? Mediators, moderators, and independent, overlapping and proxy risk factors. The American Journal of Psychiatry 2001; 158: 848–56. [DOI] [PubMed] [Google Scholar]
- Kraemer HC, Wilson GT, Fairburn CG, Agras WS. Mediators and moderators of treatment effects in randomized clinical trials. Archives of General Psychiatry 2002; 59: 877–83. [DOI] [PubMed] [Google Scholar]
- Kromrey JD, Foster‐Johnson L. Mean centering in moderated multiple regression: much ado about nothing. Educational and Psychological Measurement 1998; 58: 42–68. [Google Scholar]
- McGee D, Reed DYK. The results of logistic analyses when the variables are highly correlated: an empirical example using diet and CHD incidence. Journal of Chronic Diseases 1984; 37(9): 713–19. [DOI] [PubMed] [Google Scholar]