2
$\begingroup$

I have a data on obesity status of women in a country. This data is based on across sectional study. Now I want to find the determinants of obesity among the women. Here dependent variable is binary "obese=1/non-obese=0" based on BMI values. There are several co-variates like women age, education, occupational status, food habit, number of children, place of residence children etc. Can I use Structural Equation Model to find out the risk factors for obesity among women? or Logistic regression is the best option? Is there any other approach to do this?

$\endgroup$
  • 3
    $\begingroup$ Is there some reason why you want to code obese versus non-obese when you have actual BMI values? Much information can be lost when you dichotomize a continuous variable like BMI. You could simply have a linear regression of BMI (or perhaps some continuous transformation of BMI) against the predictor variables. $\endgroup$ – EdM Mar 27 '15 at 16:52
  • $\begingroup$ I am also using linear regression model by considering BMI (continuous variable) as response variable $\endgroup$ – Rudro88 Mar 27 '15 at 16:55
  • $\begingroup$ SEM is great at what it does, but I don't see what you need here that doesn't involved logistic regression. $\endgroup$ – Jeremy Miles Mar 27 '15 at 17:39
  • $\begingroup$ Be aware of possible confounding variables when interpreting the model, you are unlikely to pinpoint exact "determinants" with the data you have. $\endgroup$ – snoram Mar 27 '15 at 20:10
3
$\begingroup$

Logistic regression is the classical method used for such situations. Medical journal reviewers as well readers are more familiar with this method. Logistic regression will clearly shows which factors are independently associated with obesity. Odds ratio can also be calculated using logistic regression which will give an idea of the size of effect quantitatively. Most existing studies also would also have used logistic regression so the results can be compared more easily.

Methods of Structural Equation Modelling are less well known, varied, with less clear assumptions, often difficult to interpret and generally more complex.

$\endgroup$
  • $\begingroup$ Thanks. By the way, any different modelling procedure available for this? @mso $\endgroup$ – Rudro88 Mar 27 '15 at 17:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.