Is JMP predicting 0 or 1 in Logistic Regression?

[#1]

In logistic regression, the outcome has to be coded as 0 or 1. In every program I've run in SPSS, SAS, etc., logistic regression models predict the outcome (y) coded 1, relative to 0. Am I reading my output correctly that JMP is predicting the 0 outcome instead?

"For Outcome1 odds of 0 versus 1".  Shouldn't this be "1 versus 0"?

If it is 0 vs 1, how can I reverse that? I see there is an "Inverse Prediction..." option under "Nominal Logistic Fit" in the output, but that seems to just invert the predictors in the model rather than the outcome, and my output gives me the OR and the reciprocal for every predictor anyway. I don't want to have to back-calculate the odd ratios and CIs for every variable to predict the outcome I want.

This also makes me wonder how the predictors are to be interpreted. I would assume that "level1 /level2" would mean the odds of level 1 relative to level 2... but I would also assume it is 1 vs. 0 automatically.

In a multinomial logistic regression model I also ran in JMP, the model seems to be predicting Code 1 vs. Code 3, and Code 2 vs. Code 3, with the outcome coded as 1,2,3, so this makes sense. The output isn't entirely clear to me since it is very different from teh way SPSS and SAS output these models. It simply says "For log odds of value1/value3, value2/value3"

[#2 · Accepted solution]

JMP is not expecting 0 or 1 but only two nominal values that could be represented as 0 and 1, "0" and "1", "A" and "B", "Pass" and "Fail" or any other pair of binary values. This way is more general and more accommodating. It can also leading to more confusion when you get your regression results.

JMP assumes that you are targeting the 'first level.' If you enter 0 and 1, then, since 0 comes before 1 in sorting, 0 is the first level. Sometimes this alphanumeric ordering is correct but not always.

You can correct this problem and impose your own desired ordering by select the column, selecting Cols > Column Info, click Column Properties and select Value Ordering. Now select one of the levels and use the Move Up or Move Down button on the right side to achieve the desired order.

[#3]

OK, that makes sense, but is frustrating.

So, by the same token, are predictors in the model coded in the traditional 0/1 being interpreted in the same way? For example, an odds ratio of 1.5 for gender (male=1, female=0) is to be interpreted as an elevated odds for female relative to male?

[#4]

JMP will label the odds ratios so you know what it represents. You can use the same Value Ordering property to switch the order to suit your needs.

I don't understand why it is frustrating. It is different from other software but you can continue to work as before because the column property will make JMP provide the answers for any interpretation you like.

Note that you are accustomed to using 0 or 1 to record the levels ("traditional") but you don't have to use only those values. If you want to continue to use these values then just add the column property if necessary to get the order that you want. You can also use the Value Label property to make JMP display other values (such as "female" and "male")  instead of the actual values (0 and 1).