Identifying and controlling potential confounding variables is the single most important task faced by researchers. If a confounding variable is allowed to affect the results of a study, no meaningful conclusions can be drawn from the hard work of designing and running the study. Consequently, the vast majority of research design methodology is devoted to this single task.
One of the method for controlling the confounding variables is to run a multiple logistic regression. You can apply binary logistics regression if the outcome (Dependent ) variable is binary (Yes/No). In logistics regression model, under the covariates include the independent and confounding variables. The result will show the association between dependent and independent variables controlling for the confounding variable. You can use any software such as spss or stata etc.
In observational studies evaluating treatment effects, many researchers utilize "matching" type approaches. These approaches try to replicate the RCT by matching treated and untreated subjects on observed baseline characteristics. Thus, the "matched" groups are then expected to be balanced on observed pre-intervention characteristics that may otherwise confound the relationship between the treatment and the outcome (in a sufficiently large RCT unobserved characteristics are also expected to be balanced).
These "matching" approaches carry an advantage over regression-type models for analysis specifically because you can observe whether the groups are, in fact, comparable on those confounding variables. For example, if one group had a mean age of 70 and the other group had an mean age of 30 (and let's say that age is a confounder), then we can clearly see that the groups are not comparable on this variable, and the treatment effects will likely be biased. However, if instead we simply ran a regression and plugged in age as a covariate, we would get an outcome, but we would not know that there is significant extrapolation occurring in the age-space.
I am attaching links to some papers I've written on the propensity score techniques that specifically deal with this issue. I hope you find them helpful. I have also written commands in Stata that perform some of these methods, although Stata has an entire suite of commands under -teffects- that perform them as well.
Ariel
Article Combining propensity score-based stratification and weightin...
Article Combining the Regression Discontinuity Design and Propensity...
Article Designing a Prospective Study When Randomization is Not Feasible
Article Applying a propensity-score based weighting model to interru...
Article Using Propensity Score-Based Weighting in the Evaluation of ...
Article Evaluating health management programmes over time: Applicati...
Article Improving participant selection in disease management progra...
Article Using Propensity Scores to Construct Comparable Control Grou...
Hi Mohammed
By using multiple logistic regression analysis through some statistical softwares like SPSS, or Stata, Minitab, Statistica, ----etc.
This is sometimes difficult. It requires making assumptions about the models (so the propensity model papers Ariel mentions describe these).
The propensity matching procedure are one way to describes this confounding. Another relates to something called d-separation. Here is a "without tears" version.
http://bayes.cs.ucla.edu/BOOK-2K/d-sep.html
This site uses graphs to explain the assumptions. Others (e.g., Rubin) prefer not to use graph.
To rule out confounding factors, whatever the study structure, you can use adjustment tests, as X² Mantel -Haenszel, or logistic regression if it is indicated.
I agree with Faiq, i think multiple logistic regression will help; especially SPSS.
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Pharmaceuticals
- Drugs