This is concerning to v2.0.3.129 EpiData Analysis. Recently I dig in to OR and RR using said version. The data base I used was one used in SPSS to do the same. In essence, the calculation result were different from what I obtained from SPSS, given that I recoded outcomes and risk exposures, i.e. 1=2, 2=1. (SPSS uses ascending order for both outcomes and risk exposures, 1 codes as positive 2 as negative which is just the opposite of EpiData Analysis handling) EpiData Analysis result: OR=3.68 (95% CI: 1.03-11.89)Fishersexact p= 0.0224 SPSS result: OR=3.72 (3.7158) (95% CI: 1.27039-10.86604) Fishersexact p=.02239 Whereas RR resulted the same. EpiData Analysis: RR = 2.86 (95% CI: 1.31-6.24)Fishersexact p= 0.0224 SPSS: RR = 2.86 (2.85789) (95% CI: 1.30965-6.23646) USING CELL OBSERVED FREQUENCIES AS A=6, B=13, C=20, D=161 WHEREAS A:C1R1, B:C2R1, C:C1R2, D:C2R2. I wonder if the formula used by EpiData Analysis is OR=(A/C)/(B/D)=(A/B)/(C/D)= (AD)/(BC). The OR as well as corresponding CI are different from that of SPSS. The attached .dbf file uses RV3 as outcome variable and RRIESGO as risk exposure variable using code 2 as positive for both variables.
EpiData provides the conditional maximum likelihood estimate for OR, which is normally accepted as a better estimate.
If you put your data into OpenEpi you can see the differences between estimates and confidence intervals:
Jamie
epidata-list@lists.umanitoba.ca wrote:
This is concerning to v2.0.3.129 EpiData Analysis. Recently I dig in to OR and RR using said version. The data base I used was one used in SPSS to do the same. In essence, the calculation result were different from what I obtained from SPSS, given that I recoded outcomes and risk exposures, i.e. 1=2, 2=1. (SPSS uses ascending order for both outcomes and risk exposures, 1 codes as positive 2 as negative which is just the opposite of EpiData Analysis handling) EpiData Analysis result: OR=3.68 (95% CI: 1.03-11.89)Fishersexact p= 0.0224 SPSS result: OR=3.72 (3.7158) (95% CI: 1.27039-10.86604) Fishersexact p=.02239 Whereas RR resulted the same. EpiData Analysis: RR = 2.86 (95% CI: 1.31-6.24)Fishersexact p= 0.0224 SPSS: RR = 2.86 (2.85789) (95% CI: 1.30965-6.23646) USING CELL OBSERVED FREQUENCIES AS A=6, B=13, C=20, D=161 WHEREAS A:C1R1, B:C2R1, C:C1R2, D:C2R2. I wonder if the formula used by EpiData Analysis is OR=(A/C)/(B/D)=(A/B)/(C/D)= (AD)/(BC). The OR as well as corresponding CI are different from that of SPSS
I would like EpiData Analysis to state that its OR is CMLE (at least in the existing help file). AND add (regular) OR as well if you may, because both are there in OpenEpi. If you follow OpenEpi I wonder why you use Fisherexact p only instead of Mid-P Exact as well(Mid-P Exact is the preferred by the author(s) at OpenEpi by the way in this case). By the way, EpiInfo 6 Statcalc calculates OR not by CMLE and obtains same as SPSS does.
--- On Mon, 10/13/08, epidata-list@lists.umanitoba.ca epidata-list@lists.umanitoba.ca wrote:
From: epidata-list@lists.umanitoba.ca epidata-list@lists.umanitoba.ca Subject: Re: [EpiData-list] OR error report To: epidata-list@lists.umanitoba.ca Date: Monday, October 13, 2008, 7:09 PM EpiData provides the conditional maximum likelihood estimate for OR, which is normally accepted as a better estimate.
If you put your data into OpenEpi you can see the differences between estimates and confidence intervals:
Jamie
epidata-list@lists.umanitoba.ca wrote:
This is concerning to v2.0.3.129 EpiData Analysis. Recently I dig in to OR and RR using said version. The data base I used was one used in SPSS to do the
same.
In essence, the calculation result were different from
what I obtained
from SPSS, given that I recoded outcomes and risk
exposures, i.e. 1=2, 2=1.
(SPSS uses ascending order for both outcomes and risk
exposures, 1 codes as positive 2 as negative which is just the opposite of EpiData Analysis handling)
EpiData Analysis result: OR=3.68 (95% CI:
1.03-11.89)Fishersexact p= 0.0224
SPSS result: OR=3.72 (3.7158) (95% CI:
1.27039-10.86604) Fishersexact p=.02239
Whereas RR resulted the same. EpiData Analysis: RR = 2.86 (95% CI:
1.31-6.24)Fishersexact p= 0.0224
SPSS: RR = 2.86 (2.85789) (95% CI: 1.30965-6.23646) USING CELL OBSERVED FREQUENCIES AS A=6, B=13, C=20,
D=161 WHEREAS A:C1R1, B:C2R1, C:C1R2, D:C2R2.
I wonder if the formula used by EpiData Analysis is
OR=(A/C)/(B/D)=(A/B)/(C/D)= (AD)/(BC). The OR as well as corresponding CI are different from that of SPSS
EpiData-list mailing list EpiData-list@lists.umanitoba.ca http://lists.umanitoba.ca/mailman/listinfo/epidata-list
I did not mean to suggest that EpiData follows choices made in OpenEpi - that's just an easy way to see the differences between the two estimates. One of the very nice features of EpiData is that the screen is not cluttered by having multiple results. This is why you have to ask for OR or RR in the command--based on your knowledge of study design. Getting multiple odds ratio estimates AND an estimate of relative risk is confusing to many users. Why not go with the 'best' estimator rather than the one that is computationally easiest?
Mid-p estimates make sense to me (as a statistician in my past life), but they haven't gained favour in the main stream yet.
Thanks for comments - I know that the development team finds this kind of feedback useful because I have seen some of my own suggestions implemented.
Jamie
I would like EpiData Analysis to state that its OR is CMLE (at least in the existing help file). AND add (regular) OR as well if you may, because both are there in OpenEpi. If you follow OpenEpi I wonder why you use Fisherexact p only instead of Mid-P Exact as well(Mid-P Exact is the preferred by the author(s) at OpenEpi by the way in this case). By the way, EpiInfo 6 Statcalc calculates OR not by CMLE and obtains same as SPSS does.
Subject: Re: [EpiData-list] OR error report Date: Monday, October 13, 2008, 7:09 PM EpiData provides the conditional maximum likelihood estimate for OR, which is normally accepted as a better estimate.
If you put your data into OpenEpi you can see the differences between estimates and confidence intervals:
Jamie
participants (1)
-
epidata-list＠lists.umanitoba.ca