Determining which to use: Exp(est) or Odds ratio

[joesmama]

Hello,

I am trying to understand what makes the most sense in interpreting Odds Ratios for the following output. I know that the odds ratio and exp(estimate) are the same when reference groups are defined and the param=ref is used. But what about using effect coding: then the results are different .  I'm unsure of which ones to report. I'm inclined to use the ddds ratio results but need to better understand the interpretation of the effect coding on exp(est). Any clarification would be most helpful. Thank you. 

proc logistic data=temp;
class age(ref="40 and over") race (ref="White") gender(ref="M")/param=ref */param=effect*/;

model counseling  (event="Yes")=age gender race;/expb cl;

Analysis of Maximum Likelihood EstimatesParameter   DF Estimate StandardError WaldChi-Square Pr > ChiSq Exp(Est)Intercept  age <40gender FRace African AmericanRace Other   

1	-2.1228	0.7001	9.1955	0.0024	0.120
1	1.3605	0.3544	14.7401	0.0001	3.898
1	0.8608	0.3649	5.5666	0.0183	2.365
1	-0.4210	0.6152	0.4683	0.4938	0.656
1	-0.0548	0.5328	0.0106	0.9181	0.947

SAS Output

Odds Ratio EstimatesEffect Point Estimate 95% WaldConfidence Limitsage <40 vs 40 and overgender F vs MRace African American vs WhiteRacen Other vs White 

15.195	3.788	60.950
5.594	1.338	23.381
0.408	0.065	2.550
0.588	0.130	2.653

[Reeza]

Report the Odds ratio, because if you have trouble understanding it assume your audience will as well.

As to the interpretation for effect coding, even I have to write out the formula's each time I use it to verify it again so I have to look it up each time as well.

[StatDave · Accepted solution]

It is best to use the values reported in the Odds Ratios tables.  A value in the Exp(Est) column is computed by simply exponentiating the parameter estimate in that value's row. The exponentiated parameter estimate is an odds ratio only in certain cases (such as when the parameter is not involved in interactions and if it uses reference or GLM coding if it is in a categorical predictor). In other cases, the odds ratios are linear combinations of model parameters. The values in the odds ratio tables compute the appropriate linear combination of model parameters to produce the odds ratios as labeled in those tables. See "Odds ratio estimation" in the Details section of the LOGISTIC documentation for more details. 

[joesmama]

Thank you for explanation. It clarified everything!