Stata version 13 Illustration: One Way Analysis of Variance Stata version 13 Illustration One Way Analysis of Variance Page 1. Preliminary – Download anovaplot ……………………………………………………………….. 2 2. Descriptives – Graphs ………………………………………………………………………………. 3 3. Descriptives – Numerical …………………………………………………………………………… 6 4. Assessment of Normality …………………………………………………………………………….. 7 5. Analysis of Variance Model Estimation …………………………………………………………….. 9 6. Tests of Equality of Variances ………………..……………………………………………………… 11 7. Post-hoc Pairwise Comparison of Groups …………………………………………………………… 12 8. Post-hoc Graphs ………………………………………………………………………………………. 14 Source Hulley et al (1998) Randomized trial of estrogen plus progestin for secondary prevention of heart disease in postmenopausal women. The Heart and Estrogen/progestin Replacement Study. Journal of the American Medical Association, 280(7), 605-613 Source Data The actual data set contains information on n=2,763. This was a randomized controlled trial investigation of hormone therapy for the prevention of heart attack and death. Notes (1) For this illustration, I have taken a subsample of size n=612 so that students with “small stata” will be able to reproduce this analysis. (2) Specifically, I took a random sample of 300 whites, all 218 African-Americans, plus all 94 women of other race ethnicity. The total sample size of the data set used in this illustration is thus 612. Analysis Question In this hypothetical data set, does systolic blood pressure (sbp) vary by race-ethnicity (raceth)? Before you begin - Download from the course website: hers_640anova.dta …\STATA v 13\stata v 13 one way anova.docx Page 1 of 15 Stata version 13 Illustration: One Way Analysis of Variance 1. Preliminary – Download anovaplot . * Download "add-on" anova command anovaplot if you don’t already have it . findit anovaplot Click here to download Check - Download from the course website: hers_640anova.dta …\STATA v 13\stata v 13 one way anova.docx Page 2 of 15 Stata version 13 Illustration: One Way Analysis of Variance 2. Descriptives - Graphs . ** Set scheme for graphs – user choice . set scheme s1color . * get min and max of y=sbp for y-axis tick marks . tabstat sbp, stat(min max) variable | min max -------------+-------------------- sbp | 95 194 ---------------------------------- . * retrieve correspondence between codes and code labels and get sample sizes . numlabel, add . tabulate raceth race/ethnicity | Freq. Percent Cum. --------------------+----------------------------------- 1. White | 300 49.02 49.02 2. African American | 218 35.62 84.64 3. Other | 94 15.36 100.00 --------------------+----------------------------------- Total | 612 100.00 . ** Side-by-side dot plot - No frills . sort raceth . dotplot sbp, over(raceth) sbp distributions are similar in the 3 groups. It’s unlikely that anova will be significant. …\STATA v 13\stata v 13 one way anova.docx Page 3 of 15 Stata version 13 Illustration: One Way Analysis of Variance . ** Side-by-side dot plot – Aesthetics added . dotplot sbp, over(raceth) msymbol(d) msize(vsmall) mcolor(blue) ylabel(75(25)200, labsize(small)) xlabel(1 "Whites (n=300)" 2 "African-Americans (n=218)" 3 "Other (n=94)", labsize(small) angle(45)) title("Systolic BP (mm HG), by Race-Ethnicity") subtitle("n=612") caption("dot_pretty.png", size(vsmall)) …\STATA v 13\stata v 13 one way anova.docx Page 4 of 15 Stata version 13 Illustration: One Way Analysis of Variance . ** Side-by-side box plot - No frills . graph box sbp, over(raceth) . ** Side-by-side box plot – Aesthetics added . graph box sbp, over(raceth) ylabel(75(25)200, labsize(small)) title("Systolic BP (mm HG), by Race-Ethnicity") subtitle("n=612") caption("box_pretty.png", size(vsmall)) …\STATA v 13\stata v 13 one way anova.docx Page 5 of 15 Stata version 13 Illustration: One Way Analysis of Variance 3. Descriptives - Numerical . * Descriptives - Numerical Descriptives of Raw Data . tabstat sbp, by(raceth) stat(n mean sd sem min q max) Summary for variables: sbp by categories of: raceth (race/ethnicity) raceth | N mean sd se(mean) min p25 p50 p75 max -----------------+------------------------------------------------------------------------------------------ 1. White | 300 136.0133 18.55138 1.071064 98 122 133 149 189 2. African Ameri | 218 138.2339 19.99252 1.354064 98 123 137 150 194 3. Other | 94 135.1809 21.25977 2.192778 95 119 134 149 187 -----------------+------------------------------------------------------------------------------------------ Total | 612 136.6765 19.50878 .7885956 95 122 135 149 194 ------------------------------------------------------------------------------------------------------------ Numerical descriptives confirm what we saw in the graphs. The means are similar. We see that they range 135.2 mm Hg to 138.2 mm Hg. The standard deviations appear to be similar, too. They range 18.55 mm Hg to 21.26 mm Hg. Test of equality of variances is likely to be non-significant. …\STATA v 13\stata v 13 one way anova.docx Page 6 of 15 Stata version 13 Illustration: One Way Analysis of Variance 4. Assessment of Normality . * Shapiro-Wilk Test (NULL: Distribution is normal) . swilk sbp Shapiro-Wilk W test for normal data Variable | Obs W V z Prob>z -------------+-------------------------------------------------- sbp | 612 0.98098 7.681 4.945 0.00000 There is statistically significant evidence of departure from normality. Null is rejected (p < .00001). Let’s do a graphical look to see if we’re really in trouble. . ** histogram of overlay normal – No frills . histogram sbp, normal (bin=24, start=95, width=4.125) …\STATA v 13\stata v 13 one way anova.docx Page 7 of 15 Stata version 13 Illustration: One Way Analysis of Variance . ** histogram of overlay normal – Aesthetics added . histogram sbp, normal start(70) width(5) percent xlabel(75(25)200, labsize(small)) title("Distribution of Y=sbp") subtitle("Assessment of Normality") caption("histogram_pretty.png", size(vsmall)) (bin=25, start=70, width=5) This departure from normality is not so severe as to warrant transforming the data. For purposes of this illustration, we’ll proceed under the assumption of normality. …\STATA v 13\stata v 13 one way anova.docx Page 8 of 15 Stata version 13 Illustration: One Way Analysis of Variance 5. Analysis of Variance Model Estimation Stata offers at least 2 commands for a one way anova: anova or oneway. .* The command ANOVA uses deviation from means parameterization .* anova YVARIABLE FACTOR . anova sbp raceth Number of obs = 612 R-squared = 0.0037 Root MSE = 19.5042 Adj R-squared = 0.0005 Source | Partial SS df MS F Prob > F -----------+---------------------------------------------------- Model | 871.000171 2 435.500085 1.14 0.3190 | raceth | 871.000171 2 435.500085 1.14 0.3190 | Residual | 231670.941 609 380.412054 -----------+---------------------------------------------------- Total | 232541.941 611 380.592375 The null hypothesis assumption of equal means does not lead to an unlikely outcome (p=.32); the null hypothesis is NOT rejected. .* The command ONEWAY uses deviation from means and provides Bartlett test of equal variances. .* oneway YVARIABLE FACTOR . oneway sbp raceth Analysis of Variance Source SS df MS F Prob > F ------------------------------------------------------------------------ Between groups 871.000171 2 435.500085 1.14 0.3190 Within groups 231670.941 609 380.412054 ------------------------------------------------------------------------ Total 232541.941 611 380.592375 Bartlett's test for equal variances: chi2(2) = 3.1766 Prob>chi2 = 0.204 .* To obtain summary statistics, use command ONEWAY with option TABULATE . oneway sbp raceth, tabulate race/ethnic | Summary of systolic blood pressure ity | Mean Std. Dev. Freq. ------------+------------------------------------ 1. White | 136.01333 18.551379 300 2. Africa | 138.23394 19.992518 218 3. Other | 135.18085 21.259767 94 ------------+------------------------------------ Total | 136.67647 19.508777 612 Analysis of Variance Source SS df MS F Prob > F ------------------------------------------------------------------------ Between groups 871.000171 2 435.500085 1.14 0.3190 Within groups 231670.941 609 380.412054 ------------------------------------------------------------------------ Total 232541.941 611 380.592375 Bartlett's test for equal variances: chi2(2) = 3.1766 Prob>chi2 = 0.204 …\STATA v 13\stata v 13 one way anova.docx Page 9 of 15 Stata version 13 Illustration: One Way Analysis of Variance Tip! – To see the departures of the group specific means from the overall means, do a reference cell coding analysis of variance. There are multiple ways to do this. One way is to issue two commands: anova followed by regress . *** Reference Cell Coding output can be obtained with two steps: (1) anova (2) regress .* Step 1 – Command is anova . anova sbp raceth Number of obs = 612 R-squared = 0.0037 Root MSE = 19.5042 Adj R-squared = 0.0005 Source | Partial SS df MS F Prob > F -----------+---------------------------------------------------- Model | 871.000171 2 435.500085 1.14 0.3190 | raceth | 871.000171 2 435.500085 1.14 0.3190 | Residual | 231670.941 609 380.412054 -----------+---------------------------------------------------- Total | 232541.941 611 380.592375 .* Step 2 – Command is regress . regress Source | SS df MS Number of obs = 612 -------------+------------------------------ F( 2, 609) = 1.14 Model | 871.000171 2 435.500085 Prob > F = 0.3190 Residual | 231670.941 609 380.412054 R-squared = 0.0037 -------------+------------------------------ Adj R-squared = 0.0005 Total | 232541.941 611 380.592375 Root MSE = 19.504 ------------------------------------------------------------------------------ sbp | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- raceth | 2 | 2.220612 1.735814 1.28 0.201 -1.188296 5.629519 3 | -.8324823 2.305423 -0.36 0.718 -5.360027 3.695063 | _cons | 136.0133 1.126073 120.79 0.000 133.8019 138.2248 ------------------------------------------------------------------------------ Checking with means on page 9 – they match! ˆ ˆ ˆ Y = Y = [ b ] = 136.01 WHITE ref 0 ˆ ˆ ˆ ˆ Y = Y = [ b + b ] = [136.01 + 2.22] = 138.23 AFRICAN-AMERICAN 2 0 2 ˆ ˆ ˆ ˆ Y = Y = [ b +b ] = [136.01 - 0.83 ] = 135.18 OTHER 3 0 3 …\STATA v 13\stata v 13 one way anova.docx Page 10 of 15