MI estimate command, multiple imputation

[NishmaV Patel]

Hi

I have 20 imputed data sets for variableX . Patients are either in the intervention or control group.

I would like to conduct a t-test to test for a significant differences between the groups using the MI estimate command.

How can I do this? The normal command would be

ttest variableX, (by group)


However, it does not work for multiply imputed data.


Can anyone help?

Thanks in advance

[NishmaV Patel]

Hi

I have 20 imputed data sets for variableX . Patients are either in the intervention or control group.

I would like to conduct a t-test to test for a significant differences between the groups using the MI estimate command.

How can I do this? The normal command would be

ttest variableX, (by group)


However, it does not work for multiply imputed data.


Can anyone help?

Thanks in advance

[daniel klein]

You might need to give us more details, but a t-test can be expressed as a simple linear regression model. Here is an example

sysuse auto

ttest price ,by(foreign)
reg price i.foreign

Best
Daniel

[Irene Gómez]

Dear Daniel,

I am working with 50 imputed data sets. Data are from a RCT with to arm control group and intervention group. I would like to conduct a t-test to test for a significant differences between the groups using the MI estimate command. I have seen in your previous post that I can conduct a simple linear regression model.

VD= RESULT_GAD7_pos (anxiety at follow up)

Groups: CON_INT (Control or intervention group)

I have already done:

mi estimate: regress RESULT_GAD7_pos i.CON_INT

Multiple-imputation estimates Imputations = 50
Linear regression Number of obs = 3,062
Average RVI = 7.6167
Largest FMI = 0.8662
Complete DF = 3060
DF adjustment: Small sample DF: min = 57.08
avg = 166.62
max = 276.16
Model F test: Equal FMI F( 1, 276.2) = 4.93
Within VCE type: OLS Prob > F = 0.0272

-------------------------------------------------------------------------------
RESULT_GAD7~s | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------+----------------------------------------------------------------
CON_INT |
Intervencion | .4630689 .2085747 2.22 0.027 .0524707 .8736672
_cons | 3.327451 .3049485 10.91 0.000 2.716821 3.938081
-------------------------------------------------------------------------------



There is a significant difference between control and intervention condition (CON_INT) but how can I calculate mean and p values? or what do you recomend me to do?

Thank you.
Irene.

[Rich Goldstein]

first, please read and follow the FAQ which tells one, among other things, how to post so the result is easy to read

second, the p-value is shown in your output - what else do you want?

third, the coefficient for _cons is the mean for the control group (making an assumption about how you have coded things since you did not show us or tell us); the other coefficient is the difference in means - that seems to cover things (if you want the other actual mean just sum the two coefficients)

finally, note that this does not really go with this subject so you should have started a new topic with a new subject

[Irene Gómez]

Dear Goldstein,


Thank you so much for your response.

Regarding coded, control group is 1 and intervention group is 2.

When you say "the other coefficient", what do you mean? which coefficient?

Where can I see intervention mean?

Thanks in advance.

Irene.

[Carlo Lazzaro]

Irene,
as an aside to Rich's helpful advice (please read and act on FAQ recommendations,. Thanks), as per your description you compared two groups (1=control; 2=intervention; by the way: please note that it is way better to code groups - that is, levels of categorical variable - starting from 0).
The results of your OLS say that:
- the mean of the control group is 3.327451 (ie, the constant of your OLS);
- the mean of the intervention group is 3.327451+ .4630689=3.7905199.

[Irene Gómez]

Carlo, Thanks you so much.it helps me a lot.

One more question, regarding p- value would be Prob > F or P>|t|?


Kind regards,
Irene.

[Carlo Lazzaro]

Irene:
look at P>|t|.

[Irene Gómez]

Many thanks¡