JMP Odds Ratio Calculations Don't Match Mine

[#1]

I'm new to JMP (version 16.1). I'm running a binomial logisitic regression model. I have the model output set to include parameter estimates and odds ratios. When I take the logodds from the parameter estimates and exponentiate them in Excel in order to obtain the odds ratios, they don't match the odds ratio output in JMP.

My parameter estimates shows that confidence limits are Wald based, and under Odds Ratios it says that tests and confidence intervals are Wald based as well. 

Does anyone have any insight as to why my calculations differ from JMPs? How does JMP calculate the odds ratios?

Example:

JMP Output:

Term	Estimate	Std Error	ChiSquare	Prob>ChiSq	Lower 95%	Upper 95%
Intercept	2.460656	1.570863	2.45	0.1172	-0.61818	5.539491
Age	-0.03287	0.021537	2.33	0.127	-0.07508	0.009342
Sex[Female]	-0.05266	0.234692	0.05	0.8225	-0.51265	0.407328
BMI	-0.0674	0.042016	2.57	0.1087	-0.14975	0.014951
Speed	-0.01883	0.017697	1.13	0.2874	-0.05351	0.015859
Restraint All[None]	-0.43613	0.269435	2.62	0.1055	-0.96422	0.09195

My Odds Ratio Calculations: 

Term	Estimate	Odds Ratio
Intercept	2.460656	11.71249
Age	-0.03287	0.967665
Sex[Female]	-0.05266	0.948703
BMI	-0.0674	0.934823
Speed	-0.01883	0.98135
Restraint All[None]	-0.43613	0.646532

JMP's Odds Ratios

Unit Odds Ratios
Term	Odds Ratio	Lower 95%	Upper 95%	Reciprocal
Age	0.967665	0.927669	1.009386	1.033415
BMI	0.934823	0.860925	1.015063	1.069722
Speed	0.98135	0.947896	1.015985	1.019004

Range Odds Ratios
Term	Odds Ratio	Lower 95%	Upper 95%	Reciprocal
Age	0.235452	0.036752	1.508409	4.247151
BMI	0.073658	0.003042	1.783541	13.57625
Speed	0.183722	0.008099	4.167436	5.443015

Odds Ratios for Sex
Level1	/Level2	Odds Ratio	Prob>Chisq	Lower 95%	Upper 95%
Male	Female	1.111064	0.8225	0.442791	2.787911
Female	Male	0.900038	0.8225	0.358692	2.2584

Odds Ratios for Restraint All
Level1	/Level2	Odds Ratio	Prob>Chisq	Lower 95%	Upper 95%
Seatbelt	None	2.392325	0.1055	0.832019	6.878711
None	Seatbelt	0.418004	0.1055	0.145376	1.201896

Thanks in advance!

[#2 · Accepted solution]

Hi @GregPeq ,

the odds ratio table is the easiest and guaranteed way to get the correct contrasts between the categories.

the reason your calculation looks incorrect is that you have been using the "Nominal" scale for the categorical. if you use "Ordinal" you would get the exact same value as in the contrasts table. this is due to the fact that when nominal values are used JMP estimates the coefficients as difference from a non weighted average category. this makes much more sense when there are many categories. when a variable is ordinal, coefficients are estimated as the difference from the previous category.

for further information about categorical variables definitions you can follow this discussion from the past:

Let us know if it works for you,

Ron

[#3]

Thanks for your reply, @ron_horne . 

My response variable is binomial. Are you saying that I should assign an ordinal modeling type to this variable in JMP, and reserve the nominal modeling type for multinomial (non-ordinal) variables? Wouldn't this cause JMP to run an ordinal logistic regression rather than a binomial logistic regression?

Thanks,

Greg

[#4]

Hi @GregPeq ,

no, just define Sex and Restraint All as ordinal and perform the same calculation (including your own odd ratios) this should bring you to the same as JMP odds ratios.

ron

[#5]

@ron_horne, you are a gentleman and a scholar. This worked like a charm!

Thanks for your help,

Greg