Step 4: Multiple logistic regression analysis

Now it is time to perform the multiple regression analyses. With this part of the analysis we will answer the second research question:

Does the relationship persist after adjusting for year, sex, age, and teacher qualification?

Note Remember that the analyses should only be based on individuals with a value of 1 for the pop variable.

Note Remember to use factor variables when including categorical/non-binary variables.

Now we perform a multiple logistic regression analysis with conflict_dich, sex, year, age_cat, qual, and health_dich.

logistic health_dich conflict_dich i.sex i.year i.age_cat i.qual if pop==1

We can see from this analysis that the odds ratio is pretty much the same for conflict_dich (OR=6.94) as for the simple logistic regression analysis. The association for conflict_dich is statistically significant (P=0.000, CI=5.78-8.34).

Summary The relationship between conflicting demands at work and self-rated health is statistically significant. There are higher odds of reporting poor health if you have conflicting demands at work to a high extent, compared to a low extent. This association persists after accounting for sex, year, age, and teacher qualification.

Note For extra review, please re-visit this section: Multiple logistic regression.

Then we save the estimates from the model and name the model “model2.” 

estimates store model2

Step 5: Multiple logistic regression analysis including mediation

With this part of the analysis we will answer the third research question:

To what extent is the relationship explained by stress?

Note When comparing different models that are non-linear (such as logistic regression), it should be noted, before beginning, that the models are not directly comparable due to rescaling bias. This makes it a bit difficult to do mediation analysis using a comparison between the estimates of two models. However, plenty of articles use this approach, and we will also do that to answer our third research question. It can, nevertheless, be wise to be aware of the limitations of this approach on mediation. For more information, see Mediation analysis.

Now we perform a multiple logistic regression analysis with conflict_dich, sex, year, age_cat, qual, stress, and health_dich. 

logistic health_dich conflict_dich i.sex i.year i.age_cat i.qual i.stress if pop==1

From this analysis, we can see that the odds ratio for conflict_dich decreases a bit (OR=4.83) when including stress in the model. However, the association is still significant (P=0.000; CI= 3.96-5.88).

Summary When stress is also taken into account, the relationship becomes weaker (OR 4.82). This therefore means stress explains a relatively large proportion of the relationship. This relationship is statistically significant.

Then we save the estimates from the model and name the model “model3.” 

estimates store model3

Now, we create an estimate table based on the estimates we have saved from the models (model1, model2, model3). 

estimates table model1 model2 model3, p eform

Note You can display p-values and odds ratios in the estimate table by adding ‘p eform’ at the end of the command.

The relationship between conflicting demands at work and self-rated health remains and is statistically significant when adjusted for sex, year, age, and teacher qualification. When you then also adjust for stress, i.e. add stress to the other variables, then the relationship becomes weaker but is still statistically significant.