BookmarkSubscribeRSS Feed
Fluorite | Level 6

I did multiple imputations by using FCS. Now I have 5 imputed data. In my logistic regression, I have a categorical variable "systolic blood pressure" with 4 levels (0, 1, 2, 3) so I want to check the "Type 3 analysis effect" in PROC MIANALYZE to study the overall effect of systolic blood pressure. 

By using PROC MIANALYZE I only get the effect for each level, but not a global one. 

I have found someone used the TEST statement like test systolic_0, systolic_1, systolic_2/multi; But I am not sure whether that is the same as the "type 3 analysis of effect"?


Thank you for your help!


Diamond | Level 26

If you make systolic blood pressure to be categorical, then that's what SAS does (in every modeling procedure, not just MIANALYZE), it provides an estimate for each level.


If you want the "global effect", if I understand you properly, then the variable must not be categorical.

Paige Miller
Fluorite | Level 6

Thank you for your reply. But if I want to study the overall effect of systolic blood pressure (categorical) just like "Type 3 analysis of effect" table in Logistic regression. How can I do it?

Diamond | Level 26

@dandixu wrote:

Thank you for your reply. But if I want to study the overall effect of systolic blood pressure (categorical) just like "Type 3 analysis of effect" table in Logistic regression. How can I do it?

If I am understanding you properly, I don't think you can do this in PROC MIANALYZE. You could combine PROC MI with PROC GLM or PROC LOGISTIC to get a Type III analysis.

Paige Miller
Jade | Level 19

It would become a 3 step process.  PROC MI to do the multiple imputation, PROC GLM (or other analysis procedure) with a BY _imputation_ statement and an OUTPUT statement for the parameter estimates and the X'X inverse matrix, and PROC MIANALYZE to tie all of this together.  See 

Example 76.6 Reading GLM Results from PARMS= and XPXI= Data Sets

in the PROC MIANALYZE documentation.



SAS Employee

There isn't an easy way to get a combined test analogous to Type3 tests in Proc MIANALYZE.  My suggestion would be to instead combine the Chi-Square statistics associated with the Type3 tests in LOGISTIC using Dr. Paul Allison's %COMBCHI macro.

Rhodochrosite | Level 12 lvm
Rhodochrosite | Level 12

I agree with SAS_Rob. Allison's %COMBCHI macro is excellent and easy to use. I have put it to great use.  All you need is the chi-squared statistic and df from each of the imputations, and it uses a method by Li et al. (1991, I think), to construct an F test for global significance. You get an overall p-value as well.

Fluorite | Level 6
Hello SAS_Rob,

Thank you for your help. I found that solution one month ago and I tried.
But it didn't work for one of my variables. It didn't give a p-value. And I
am also not clear that what is the difference between this F-test for
global significance and the Wald chi-square test used in "type three
analysis of effect".

I came up with a solution to calculate the combined wald test. After I get
the pooled estimation for each level. I used the equation to calculate the
chi-square: transposed estimation matrix*inverse matrix of covariance
matrix* estimation matrix.

Quartz | Level 8

The link given to the %COMBCHI macro is not being updated and still points to v1.0. There is an amended formula in version 2.1 of  Dr. Paul Allison's %COMBCHI macro from 2007 linked from the "Resources" section of his bio at


Don't miss out on SAS Innovate - Register now for the FREE Livestream!

Can't make it to Vegas? No problem! Watch our general sessions LIVE or on-demand starting April 17th. Hear from SAS execs, best-selling author Adam Grant, Hot Ones host Sean Evans, top tech journalist Kara Swisher, AI expert Cassie Kozyrkov, and the mind-blowing dance crew iLuminate! Plus, get access to over 20 breakout sessions.


Register now!

What is ANOVA?

ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 8 replies
  • 1 like
  • 6 in conversation