[U] User's Guide

292 [ U ] 20 Estimation and postestimation commands. margins sex#agegrp, at(bmi=(20 30))Adjusted predictions Number of obs = 10351Model VCE : LinearizedExpression: Pr(highbp), predict()1._at : bmi = 202._at : bmi = 30Delta-methodMargin Std. Err. z P>|z| [95% Conf. Interval]_at#sex#agegrp1 1 1 .0109371 .0032537 3.36 0.001 .00456 .01731421 1 2 .0156812 .0047009 3.34 0.001 .0064677 .02489471 1 3 .0482213 .0122271 3.94 0.000 .0242566 .0721861 1 4 .0949467 .0173769 5.46 0.000 .0608887 .12900481 1 5 .101735 .0142451 7.14 0.000 .0738152 .12965481 1 6 .108137 .0203938 5.30 0.000 .0681659 .14810821 2 1 .0053248 .0022848 2.33 0.020 .0008466 .00980291 2 2 .0101007 .0036738 2.75 0.006 .0029001 .01730121 2 3 .0422142 .0118314 3.57 0.000 .0190251 .06540331 2 4 .0912119 .0201067 4.54 0.000 .0518035 .13062031 2 5 .0927126 .011551 8.03 0.000 .070073 .11535211 2 6 .1035166 .0235733 4.39 0.000 .0573138 .14971942 1 1 .095307 .018967 5.02 0.000 .0581324 .13248152 1 2 .1218371 .0174947 6.96 0.000 .0875481 .15612612 1 3 .2275182 .027168 8.37 0.000 .17427 .28076652 1 4 .3084116 .0306717 10.06 0.000 .248296 .36852712 1 5 .3276512 .0235645 13.90 0.000 .2814657 .37383672 1 6 .2001169 .0352892 5.67 0.000 .1309514 .26928252 2 1 .0258852 .007303 3.54 0.000 .0115715 .04019892 2 2 .0483494 .0092976 5.20 0.000 .0301264 .06657232 2 3 .1303522 .0198102 6.58 0.000 .0915249 .16917952 2 4 .1946575 .0192303 10.12 0.000 .1569668 .23234822 2 5 .2338165 .0167405 13.97 0.000 .2010058 .26662722 2 6 .2582884 .0256129 10.08 0.000 .2080881 .3084887That is a lot of margins, but they are in sets of six age groups. The first six margins are menwith a BMI of 20, the second six are women with a BMI of 20, the third six are men with a BMIof 30, and the last six are women with a BMI of 30. These margins tell a more complete story. Theprobability of high blood pressure is much lower for both men and women who maintain a BMI of 20.More interesting is that the relationship between men and women differs depending on BMI. Whileyoung men who maintain a BMI of 20 are still twice as likely as young women to have high bloodpressure (0.011/0.005) and youngish men are over 50% more likely (0.016/0.010), the gap narrowssubstantially for men in the four older groups. The story is worse for those with a BMI of 30. Bothmen and women with a high BMI have a substantially increased risk of high blood pressure, with menages 50–69 almost 10 percentage points higher than women. Before you dismiss these differences ascaused by the usual attenuation of the logistic curve in the tails, recall that when we fit the model,we allowed the effect of bmi to be different for each combination of sex and agegrp.You may have noticed that the header of the prior results says “Adjusted predictions” rather than“Predictive margins”. That is because our model has only three covariates, and we have fixed thevalues of each. margins is no longer averaging over the data, but is instead evaluating the marginsat fixed points that we have requested. It lets us know that by changing the header.We could post the results of margins and form linear combinations or perform tests about any ofthe assertions above; see Example 10: Testing margins in [R] margins.