plot a 95% confidence interval in a logistic regression

[boban]

Hi there,

I would like to add 95% confidence interval bands to the following code.

proc sgplot data=AllStats noautolegend;

axis1 label = (font = 'arial');

axis2 label = (font = 'arial');

label predmupa = 'Hatching probability'

jsh = 'JSH';

series y=predmupa x=jsh;

scatter y=predmupa x=jsh;

Does anyone know how to do this? I used proc glimmix to produce the predicted probabilities.

Thanks in advance!

Bob

run;

[Reeza]

Axis statements do no apply to SG procedures, use XAXIS instead, but the syntax is different.

Do you have the upper and lower estimates for your band? I know that PROC LOGISTIC generates these but not sure about GLIMMIX. If you have the values, a BAND statement is what you’re looking for to create a band. Order does matter here, make sure it’s before the SERIES statement.

[boban]

Thank you for your reply!

I don't have the estimates for the bands, and don't know how to get them. Is there a way of obtaining these bands, or plot in any other way?

Best,

Bob

[RW9]

You can get confidence intervals from a number of procedures depending on what you need - not really an expert, a statistician would be best to ask (proc ttest, means etc.).  You may be able to do it directly in the graph:

http://support.sas.com/documentation/cdl/en/grstatproc/62603/HTML/default/viewer.htm#a003155517.htm

[boban]

Thanks!

I tried this but it doesn't seem to work. l'll keep looking.

Cheers,

Bob

[boban]

I think I solved it!

I used this code:

proc sgplot data=AllStats noautolegend;

axis1 label = (font = 'arial');

axis2 label = (font = 'arial');

label predmupa = 'Hatching probability'

jsh = 'JSH';

band x = jsh lower = lclmupa upper = uclmupa/legendlabel = "95% clm" name ="band1";

series y=predmupa x=jsh;

scatter y=predmupa x=jsh;

keylegend "band1"/ location = inside position = bottomright;

title;

run;

I guess this will plot a 95% confidence interval around the predicted mean. Is that right?

Bob

[RW9]

That depends on how lclmupa and uclmupa were calcualted in the dataset allstats.  If you used proc means, then there is an option called alpha which you can set to be most values, but 0.05 from memory give 95% CI.

[Reeza]

In this respect, the bands need to be calculated from the GLIMMIX procedure and then you use the BAND statement. 

@boban only you can know if its correct. We can't see your data or log or graph or Glimmix code or know what you're trying to do beyond what's posted here. 

[boban]

I didn't calculate the bands. I simply requested all stats to be computed from my original data set, and used the out option to save them under the new data set (named all stats).

I then used predmupa to get the predicted probabilities ploted in proc sgplot.

Thanks for your help!

Bob