#### AndyRob1995

##### New Member
Hello,

I have also posted in Calculating adjusted odds ratio - Cross Validated (stackexchange.com)

I am in the process of teaching myself about odds ratios, adjusted odds ratios and modelling with regression for my MSc.

I am working with various publications but I have a question about how the statistics are calculated in the publication below (I am NOT challenging the calculations – just trying to follow how they are done)

Post‐disaster physical symptoms of firefighters and police officers: Role of types of exposure and post‐traumatic stress symptoms (wiley.com)

I am able to work out the unadjusted odds ratio in the table (Table 2) below (from the pub above).
However, looking at the adjusted odds ratio, it is adjusted for background characteristics. The background characteristics are in the table 2 down (Table 1) (from the pub above).

My questions are:
• Am I correct in saying that I can’t calculate the adjusted odds ratio using the background characteristics given? eg I would need the Somatic symptoms split for male/female etc?
• Am I correct in saying that that each adjusted odds ratio is adjusted for more than one background characteristic at once?
• Is this related to modelling (ie forwards and backwards) in any way.
I have tried to find some worked examples but am struggling to find any. Any help would be most appreciated.

Last edited:

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Have you conducted logistic regression before. Once familiarizing yourself with it - most of these questions will be resolved. Regardless of what software you ended up using the "Logistic Regression Using SAS" by Paul Allison is a very approachable entry level book that if you can't find online is probably fairly cheap.

Logistic regression, an approach when the dependent variable (DV) is binary.

There is simple logistic regression: one independent variable (IV), AKA predictor and multiple logistic regression (multiple IVs in model).

When someone says they adjusted for covariates you end-up with adjusted odds ratios (AOR), and it is usually implied they are adjusted for all of the covariates of interest at once.

When building a model one should use context knowledge about the phenomenon under investigation and known relationship between variables, not forward or backward stepwise selection. So if you see that a person used that approach it is a red flag they may not know what they are doing or that their results may have issues. Also, it is a faux pas to build a model and report estimates from it as though they are generalizable - since those data were used during the formative process. So if someone does this as well it is a red flag for results to have generalizability issues.

#### AndyRob1995

##### New Member
Have you conducted logistic regression before. Once familiarizing yourself with it - most of these questions will be resolved. Regardless of what software you ended up using the "Logistic Regression Using SAS" by Paul Allison is a very approachable entry level book that if you can't find online is probably fairly cheap.

Logistic regression, an approach when the dependent variable (DV) is binary.

There is simple logistic regression: one independent variable (IV), AKA predictor and multiple logistic regression (multiple IVs in model).

When someone says they adjusted for covariates you end-up with adjusted odds ratios (AOR), and it is usually implied they are adjusted for all of the covariates of interest at once.

When building a model one should use context knowledge about the phenomenon under investigation and known relationship between variables, not forward or backward stepwise selection. So if you see that a person used that approach it is a red flag they may not know what they are doing or that their results may have issues. Also, it is a faux pas to build a model and report estimates from it as though they are generalizable - since those data were used during the formative process. So if someone does this as well it is a red flag for results to have generalizability issues.