Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "Seed, Paul" <paul.seed@kcl.ac.uk> |
To | "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu> |
Subject | st: Adjust versus margins following logistic regression |
Date | Thu, 16 Dec 2010 22:32:29 +0000 |
Dear Statalist, Why does StataCorp prefer -margins- over -adjust-? I am currently using version 11.1 in Windows, and am already supposed to use version control. Is there a risk we will lose -adjust- in future versions? I recently compared -adjust- to -margins- for the estimation of adjusted percentages following logistic regression. The Stata help suggests that "adjust has been superseded by margins. Except for adjust's generate() and stdf options, the margins command can do everything that adjust did and more." But this does not seem to be the case: -adjust- respected the variable & value labels, produced valid estimates based on asymmetric confidence intervals, and allowed me to select the columns output and choose the format. -margins- did none of these things. The table from -adjust- is close to publication quality (after converting proportions to percentages and tidying in a Word processor), and could be used for an internal report as it stands. The table from -margins- is useless. Both content and formatting are wrong; and one given value is impossible. The failure to respect labels is a general problem with factor variable handling. Neither -xi- nor factor variables do well. It would be a great pity to lose one command of the few commands (-tabulate- is another) that gets it right. Best wishes, Paul Seed The commands were: xi: logistic vbac i.bmi_cat i.parity_cat adjust _Iparity_ca_2 _Iparity_ca_3, by(bmi_cat) pr ci format(%4.2f) logistic vbac i.bmi_cat i.parity_cat margins i.bmi_cat This gave two tables: ---------------------------------------------- BMI | (kg/m2) | pr lb ub ----------+----------------------------------- <20 | 0.35 [0.09 0.75] 20-25 | 0.39 [0.30 0.50] 26-30 | 0.49 [0.37 0.62] 31+ | 0.28 [0.17 0.44] ---------------------------------------------- Key: pr = Probability [lb , ub] = [95% Confidence Interval] Predictive margins Number of obs = 202 Model VCE : OIM Expression : Pr(vbac), predict() ------------------------------------------------------------------------------ | Delta-method | Margin Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- bmi_cat | 1 | .354512 .1962361 1.81 0.071 -.0301037 .7391277 2 | .3956257 .0509108 7.77 0.000 .2958423 .4954091 3 | .494516 .0636103 7.77 0.000 .3698422 .6191898 4 | .2863436 .0700777 4.09 0.000 .1489939 .4236933 ------------------------------------------------------------------------------ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/