Logistic regression parameter estimates opposite of expected

[#1]

Hello All,

I am a JMP newbie and to learn about JMP, I've been re-analyzing some of my old data using JMP. When I run a logisitic regression model with an ordinal dependent variable (0 or 1) and a continuous variable as the independent variable (a number between -0.5 and 0) the parameter estimates I get are numerically opposite to what I would expect. The parameter estimate for the independent variable (X) is -1.513 but when I run the same analysis using SAS, EpiInfo, Mac StatPlus, or R, I get 1.513. The p-value is the same and the intercept parameter estimate is also opposite. What am I missing to get JMP to work properly? I'm still using the free trial period to figure out whether I want to buy JMP or not... Thanks.

Nick

[#2]

Any chance you reversed the 0's and 1's when creating the JMP file?

[#3]

Hi,

I am also having the same issue...

In my dataset, the outcome variable, "madeit", is nominal, Yes/No and "FullDayCount" is continuous. When running Fit model - Logistic regression JMP seems to express Log odds of No versus Yes, instead of Yes versus No. Is there a way to reverse it? The stata result is correct and opposite of what JMP shows - please see attached images.

Thank you for helping me figure this out.

Patrick

JMP 10 windows version:

Stata 11 Mac:

[#4]

JMP treats the levels in alphabetic order by default. You can either recode the levels (e.g. No=0, Yeas=1) or set the Value Ordering property of the column MadeIt. To do the latter, just chose Value Ordering from the properties in the column info dialog and move Yes to the position above No. This will not only affect the odds ratio definition but also the plot order. The curve will be reversed.

[#5]

Thank you very much! Your suggestion fixed my issue.