Odds ratio or Crude Odds ratio are obtained when you are considering the effect of only one predictor variable ie your equation consists of only one independent variable. However when you include more variables in the analysis (confounder variables for the said relationship) you get what is called and Adjusted Odds Ratio, which takes into account the effect due to all the additional variables included in the analysis. 

Thank you, Ritual Kamal !! Why is it needed ? What is the application and philosophy of adding more variables and then seeing effect !!

It depends on your research topic and available variables. Unadjusted or crude ORs means that you are not looking at any other factor, but this is not realistic. Other factors associate with the dependent variable and that's why we need to add them and "adjust" the model.

If you calculate your odds ratio using bivariate analysis,  say chi square test for example that is crude odds ratio because the effects of confounders have not been controlled for. If you calculate the odds ratio using multivariate analysis to look at effects of individual factor that is adjusted odds ratio.

@Fajri...You can use logistic regression directly however, using chi sqaure 1st will reduce the number of independent variables you will include in your model; in that case you can do forward regression which considers only factors that are significant. 

forward regression will lead you directly and quickly to the refined results, free from confounding effects.

@Adamu I think Unless you change the responses of multinomial responses to dicotumus we can not use chi-square to calculate the odds ratio so, chi-square used for any type of data to test presence of reationship. but to calculate COR we have to use cross tabulation with 2x2 table.

I read the answer given by Ritual Kamal as follows

" Odds ratio or Crude Odds ratio are obtained when you are considering the effect of only one predictor variable ie your equation consists of only one independent variable. However when you include more variables in the analysis (confounder variables for the said relationship) you get what is called and Adjusted Odds Ratio, which takes into account the effect due to all the additional variables included in the analysis."

However as I have observed most researches and noble data analyzers are in trouble how to calculate the AOR. The Curde odd ratio can easisly be calculated using SPSS in the cross tabulated options. In the cross tabulated case you cannot make analysis that have more than two options in any of the variables and at the same time you cannot compute the AOR using cross tabs. Therefore you need to conduct further analysis in the logit-probit or logistic regression analysis. In this options you can involve the both discreet and continuous variable as well more number of confounding variables to conduct the AOR. It is called adjusted due to the fact the OR is computed by adjusting based on the confounding variables.

Thank you

It was a nice question. Here you will get explanation of odds ratio.

Article Explaining Odds Ratio

Logic behind using odds ratio is to check if the factor we consider is a risk factor or protective factor.

Try to explain it with an example: You want to check is mother age is a risk factor for child mortality so your dependent (main) variable is child mortality (yes or no) and independent variable is mother's age (18, 20,....).

In that case if you use binary logistic for these two variables it will give you a odds ratio (OR). Then value of the OR shows you either it is risk or protective factor.

In addition if you add another variable like income status as independent variable with previous one the your OR converted to income adjusted OR or adjusted OR. You can add many more independent variables in your analysis based on previous literatures.

Note: Before doing multivariate analysis check variables association in bivariate analysis.

https://www.sagepub.com/sites/default/files/upm-binaries/6417_Chapter_1__Meyers_I_Proof.pdf

Article A Purposeful Selection of Variables Macro for Logistic Regression

Article Appropriate Uses of Multivariate Analysis

What does this statement mean 'backwards stepwise approach used to retain variables with P < 0.1 but reported those with

Backward stepwise approach in logistic regression includes all the variables at the beginning of the analysis but eliminates the least useful predictor variable one by one (one at a time) in order to choose the variable(s) that give the exact or appropriate model of interest. The sentence implies that you remove any variable whose P value is > 0.1 (this value could differ from study to study) i.e. they should not be included in the backward stepwise logistic analysis (include only those with P

Is odds ratio that we get from calculating binary logistic regression is in itself adjusted odds ratio .

This one is possibly ancient, but I'm going to leave it here:

1) First and foremost, "Adjusted OR" (AOR, aOR, adjOR, and so on) is NOT a statistical method! (as some continue to believe...) It's simply a phrase for a result, similar to "crude OR," that's used in reports and studies in several domains of biostatistics when reporting logistic regression results.

They are, in fact, just labels for the reported findings of multi-variate ("adjusted") and uni-variate ("crude") logistic regression, respectively.

2) Even if it is common knowledge (oh, well...) that the "Adjusted OR" is achieved by re-running the logistic regression* with the same dependent variable and multiple independent variables**, there is no "standard procedure" for this, and it is entirely up to the researcher's discretion - and skill.

* having all the conditions (prerequisites) met

** Which variables to put in this soup, is another good question:

Typically, some people run a uni-variate LR with all dependent variables believed to have some potential utility, then pick only those that indicate a stat - significant level below the threshold (eg p0.05) and put them in a second LR to determine the "aOR."

Others, run a selection method (Lasso, stepwise, an "all-in" regression, etc) and select all the variables below a more relaxed threshold (eg p< 0.1) and put it in the new regression to find the aOR.

Finally, the pedant tests the independent variables in pairs and groups, attempting to comprehend their meaning and interaction before selecting only the meaningful. Of course, that's a grueling task!

Conclusion: Keep in mind that the results of the univariate, bivariate, and multivariate LRs may generate different coefficients (or Odd Ratios) for the same variable, regardless of how you do it! In most cases, an indicator's effect on the dependent variable is "reduced" by interactions with other factors rather than being produced solely by the indicator.

Sadly, some fancy explanatory variables taken alone will become not significant in a multivariate model. Furthermore, interactions might be difficult to explain, and the delicate influence of variables/confounders necessitates additional (painful) investigation. Also, there's a matter of time.

Last but not least, statistical significance does not imply that there is a causal relationship between variables. And this is a common problem in analytical models like these.

Congratulations & Good luck!

Just to add some questions raised.on OR VS AOR

When run binary logistic regression get EXp(B) value or ODD RATIO , you may get singficant p value

In this case our conclusion will be the regression is significat at binary level but not in multivarate level.

I am working with eighteen independent variables (IV). In the chi-square test, I found eight IVs significant (p-

Hello, I need a do file for running logistics regression, reporting odd ratio, crude ratio and risk ratio for stats

What is the interpretation when OR and aOR are 1.03

@Z.Asgari. Adjusted Odds ratios: adjust the OR (crude OR) for confounders or effect modifiers.

OR is used for calculating 2x2 for one variable independent vs one variable dependent. AOR used for more than one independent vs one variable dependent.

@Md Abdul. Take the AOR results for your discussion because it shows the results after controlling confounders. Its results are the real determinants of the dependent variable.

The variability of a measured varaibel could not fully explained by in factor. The crude OR is computed ignoring the influences of others factored. However, it is natural there are more factors influencing it. In this case adjusted OR becomes into picture to account this.

Great response