|Date: ||Tue, 29 Aug 2006 11:05:40 -0400|
|Reply-To: ||Lou <charl_bean@YAHOO.CO.UK>|
|Sender: ||"SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>|
|From: ||Lou <charl_bean@YAHOO.CO.UK>|
|Subject: ||Re: Compare odds ratios|
|Content-Type: ||text/plain; charset=ISO-8859-1|
While I'm on the topic of odds ratios, just another quick question. I
have calculated both unadjusted and adjusted odds ratios for the three
independent variables in my logistic regression model (age, ethnicity,
deprivation). I also have 95% CIs. For the age and deprivation
variables, the unadjusted and adjusted odds ratios are practically the
same. In the case of the ethnicity variable, however, there are some
fairly large differences, with the adjusted odds being larger (but still
less than 1). Does this mean that the influence of ethnicity on screening
is not as large as the influence of deprivation and age and that adjuting
for age and deprivation shows the reduced effect of ethnicity?
On Tue, 29 Aug 2006 10:55:34 -0400, Lou <charl_bean@YAHOO.CO.UK> wrote:
>Thanks for your response. I think it's fair to say that I'm having a bad
>day (month, actually) and my brain packed up for a while there. I
>realised as soon as I posted the question that it was stupid! I
>understand everything you have said below, so feel happy at interpreting
>the results. I think I'm going to have to adopt a policy of not looking
>at this other report because it sends me insane!!
>Thanks to you and everyone else who responded for your help.
>On Tue, 29 Aug 2006 16:50:27 +0200, =?ISO-8859-1?B?
>TWFydGEgR2FyY+1hLUdyYW5lcm8=?= <firstname.lastname@example.org> wrote:
>>If you have used logistic regression, then the answer is right in the
>>- Check the dummy coding table to find out which dummy variable of the
>>set reflects "Muslim group" (could be ethnicity(1), or ethnicity(2)...
>>depending on the order of the categories).
>>- Check the Wald test significance for that dummy variable: that's the
>>significance you are looking for (caveats: not very sensitive, but it
>>can't be helped). Alternatively, the same information can be deduced
>>from the 95%CI for exp(b), provided you've been farsighted enough to
>>ask for them (this reminds me I have to write to Kyle Weeks
>>concerning some ideas and suggestions for SPSS 16, and that's one of
>>them: 95%CI should be default output in most statistical methods).
>>If you get lost with this explanation (about locating dummies and Wald
>>test and so on), just send the logistic regression output (draft) and
>>I'll point the important results to you
>>The term "adjusted odds-ratio" the authors used in the report means
>>that they used multiple logistic regression to control for other
>>factors (age and deprivation in this case).
>>L> Okay, I'll try to clarify matters. As usual, I'm trying to immitate
>>L> that has been done previously, so the whole thing may be
>>The old "imitation theorem": "if it's published, then it's correct" ;)
>>L> I have used logistic regression to obtain adjusted odds ratios. The
>>L> outcome is 'screened' or 'not screened' and the independent variables
>>L> age, ethnicity and deprivation. I am focusing on the odds ratios for
>>L> ethnicity variable. When I talk about reference group, I mean that
>>L> odds ratios are all calculated with respect to one group. In other
>>L> I have five Asian categories (Hindu, Muslim, Sikh etc) and the odds of
>>L> undertaking screening are compared to the odds for the non-Asian group
>>L> (the reference group). I am trying to show that the odds of
>>L> screening are lower for the Asian groups when compared to the non-
>>L> In previous analysis, the people who produced the report have stated
>>L> following, "the Muslim group demonstrated a significantly (p<0.05)
>>L> uptake with adjusted odds ratio 0.37 versus 1.0 for non-Asians". I'm
>>L> basically trying to work out what they did and assumed they somehow
>>L> compared the odds.
>>L> Hope this clarifies things a bit - any suggestions gratefully
>>L> Best wishes,
>>L> On Tue, 29 Aug 2006 16:24:52 +0200, =?ISO-8859-15?B?
>>L> TWFydGEgR2FyY+1hLUdyYW5lcm8=?= <email@example.com> wrote:
>>>>Perhaps you havent' explained yourself completely... or I see an error
>>>>in your question: you just can't compare and odds-ratio with an odd,
>>>>and an OR from a group with the one of its reference group. I think
>>>>I'm being confused by the use of the term "reference group". Are you
>>>>talking about one qualitative variable that has been dummy coded, with
>>>>one group being the reference or is it something different?
>>>>L> Could someone please tell me the correct test to use to compare
>>>>L> two odds ratios are significantly different. For instance, I want
>>>>L> compare the odds ratio for group A (odds ratio = 0.487) against the
>>>>L> ratio for the reference group (odds = 1).
>>>>Dr. Marta García-Granero,PhD mailto:firstname.lastname@example.org
>>>>"It is unwise to use a statistical procedure whose use one does
>>>>not understand. SPSS syntax guide cannot supply this knowledge, and it
>>>>is certainly no substitute for the basic understanding of statistics
>>>>and statistical thinking that is essential for the wise choice of
>>>>methods and the correct interpretation of their results".
>>>>(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)
>>L> __________ Información de NOD32, revisión 1.1724 (20060824) __________
>>L> Este mensaje ha sido analizado con NOD32 antivirus system
>>Dr. Marta García-Granero,PhD mailto:email@example.com
>>"It is unwise to use a statistical procedure whose use one does
>>not understand. SPSS syntax guide cannot supply this knowledge, and it
>>is certainly no substitute for the basic understanding of statistics
>>and statistical thinking that is essential for the wise choice of
>>methods and the correct interpretation of their results".
>>(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)