e-Articles (Chapter 15)
Here are some full-length research articles that illustrate the use of the analysis of covariance. To view any article, simply click on its title. (NOTE: No claim is made that these articles are perfect in all respects. By carefully reviewing them, you will hone your skills at being able both to decipher and to critique statistically-based research reports.)
Although the researchers used intact groups in this study, their discussion of ANCOVA's assumptions is highly admirable. To be more specific, the researchers attended to the assumptions (a) that the relationship between the covariate and the dependent variable should be linear, (b) that the treatment should not affect the covariate, and (c) that the regression slopes should be homogeneous. See sections 6.1, 6.2, and 6.3.
In this randomized experiment, a one-way ANCOVA was used to compare two groups of students. This was done twice, once for each of the study's 2 dependent variables. For each analysis, the ANCOVA summary table is presented. The researchers deserve high marks for focusing their attention on adjusted means produced by each analysis. See the sections of the article called "Analysis of Data" and "Findings," as well as Tables 4 and 5.
In this study's randomized experiment, 2 nutrient supplements were compared in terms of their effects on body characteristics (e.g., % fat). Several one-way ANCOVAs were conducted, each focused on a different dependent variable, and each using baseline data as the covariate. In the research report's "Conclusion," the researchers state that: "The use of SOmaxP [one of the 2 supplements] four times per week for nine weeks resulted in statistically significant improvements in [2 measures of] strength, muscle endurance, lean muscle mass, and percentage body fat versus a comparator with identical quantities of creatine, whey protein and carbohydrate." If the Bonferroni adjustment had been used to deal with an inflated Type I error risk, 4 of the 5 significant finding would have vanished!
Illustrates the use of a 2x3 mixed ANCOVA and a 2x9 mixed ANCOVA, with attention paid to the assumption that the population within-group regression slopes should be the same. Table 5 displays both the unadjusted and adjusted means for the 2x9 ANCOVA.
Copyright © 2012
Schuyler W. Huck