Abstract
At the end of the previous chapter, we spent some time considering the difference between statistical significance and the strength of an association. In this chapter, we spend some time on a measure that combines elements of both: the odds ratio, which for 2 ×2 tables, is the recommended test of both the strength and the statistical significance of an association. As with many SAS statistical procedures we have seen, the actual execution of the procedure is fairly simple. The challenge is in understanding the underlying statistical process and interpreting the results.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Note that this does not mean that the probability is 1/3 for or against.
- 2.
We used to teach the “rare disease assumption” as the reason that a case-control study can approximate a cohort study. There are now more fundamentally sound reasons supporting the validity of case-control studies. Rothman, in particular, has a very nice discussion. Still, the reason the odds ratio is a valid approximation to the relative risk is because the outcomes we study are, in fact, rare and do not when removed detract appreciably from the bottom number of the ratios themselves when compared to the denominator of the analogous probability. (See the formula for converting probabilities to odds.) Odds ratios will overestimate risk, when the outcomes are common. As a general rule of thumb, I find outcomes in much beyond 10 to 15 percent of the study population to be problematic.
- 3.
Smoking is so frequently a confounder that you might want to at least consider including it in almost any study in which it is not the actual exposure of interest.
- 4.
You also get the same two types of estimates for cohort data.
- 5.
You can also request / CMH rather than / all. If there is a 2 ×2 table, SAS will return the Mantel–Haenszel statistic in addition to the Cochrane–Mantel–Haenszel statistics.
- 6.
As you start to stratify by more than one variable, you may find that you will start having problems with small numbers. SAS has options to request exact tests for odds ratios. You request them with an “exact” statement on its own line.
- 7.
In fact, you wont go too far wrong just using type three routinely.
- 8.
There are no hard and fast rules for what constitutes an “appreciable” or “meaningful” difference between a crude and an adjusted estimate, nor is the comparison amenable to statistical testing. This is one of those areas (again) where training, experience and substance matter knowledge combine to guide our efforts. A 15% or 20% difference is an acceptable rule of thumb.
- 9.
Of course, the sample size was pretty small, and even with Tarone adjustment we may have to question these results. Also, see chapter 11 for a discussion of an epidemiological approach to interaction. You would want to consider using the Darroch/Rothman approach.
References
Cavalieri P, Marovich P, Patetta MJ, Walsh S, Bond C, SAS Institute (2000) Statistics I: Introduction to anova, regression, and logistic regression: course notes. SAS Institute, Cary, NC
Daniel WW (2006) Biostatistics: a foundation for analysis in the health sciences 8th edition with SPSS software CD Rom 14.0 set (Wiley series in probability and statistics). Wiley, New York
Darroch J. (1997) Biologic synergism and parallelism. American Journal of Epidemiology Vol 145, No7, pp 661–668.
Delwiche LD, Slaughter SJ (2008) The little SAS book: a primer. SAS Institute, Cary, NC
Hennekens CH, Buring JE, Mayrent SL (1987) Epidemiology in medicine. Lippincott Williamns & Wilkins, Philadelphia
Hosmer DW, Lemeshow S (2000) Applied logistic regression (Wiley series in probability and statistics). Wiley, New York
Kelsey JL, Whittemore AS, Evans AS, Thompson WD (1996) Methods in observational epidemiology. Oxford University Press, Oxford, New York
Kleinbaum DG, Kupper LL, Nizam A, Muller KE (2007) Applied regression analysis and multivariable methods (Duxbury applied). Duxbury Press, North Scituate, MA
Patetta MJ, Amrhein J (2005) Categorical data analysis using logistic regression: course notes. SAS Institute, Cary, NC
Rothman KJ, Greenland S, Lash TL (2008) Modern epidemiology, 3rd Edition. Lippincott Williamns & Wilkins, Philadelphia
Schlesselman JJ. (1982) Case-control studies: design, conduct, analysis. Oxford University Press, Oxford, New York
Susser ES, Schwartz S, Morabia A, Bromet E (2006) Psychiatric epidemiology: searching for the causes of mental disorders. Oxford University Press, Oxford, New York
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2013 Springer Science+Business Media New York
About this chapter
Cite this chapter
DiMaggio, C. (2013). Categorical Data Analysis II. In: SAS for Epidemiologists. Springer, New York, NY. https://doi.org/10.1007/978-1-4614-4854-9_9
Download citation
DOI: https://doi.org/10.1007/978-1-4614-4854-9_9
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4614-4853-2
Online ISBN: 978-1-4614-4854-9
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)