Introductory note
This eBook is based on a section of a short course on concepts and methods in causal inference run by the PATHWAYS node. The original practical, on mediation, asked participants to fit models using Stata themselves, whereas this eBook fits models in Stata on the user's behalf. You therefore have to have access to Stata, and to tell Stat-JR where it is: see the third-party software page on the Stat-JR site for more information on how to specify the path to Stata and other software packages.
Question
Causal diagram
Body mass index is known to affect the level of gamma-glutamyl transpeptidase, but is also potentially affected by alcohol consumption. Socio-economic position is potentially a common cause of alcohol consumption, body mass index and systolic blood pressure.
For simplicity let's assume that the following diagram represents the relationships between the variables (where BMI = body mass index, GGT = gamma-glutamyl transpeptidase, SEP = socio-economic position, and SBP = systolic blood pressure):
Data
We will use a simulated dataset, inspired by the population-based controls from a case-control study (the Izhevsk Family Study), which was carried out to assess the effects of hazardous alcohol-drinking on premature mortality in men living in Izhevsk, Russia.
Here are the variables:
Please wait a moment whilst the dataset summary loads...
Traditional mediation analysis
First, we will perform a traditional mediation analysis, as described, for example, in the PATHWAYS workshop 'Current and new approaches to estimating causal pathways from observational data'.
What you need to do
Below you'll see some quiz questions...
- Read through these, and think about what sort of models you need to fit to answer them.
- Then, underneath the quiz questions, you'll see an input box allowing you to specify your response variables and explanatory variables to fit a linear regression model.
- Once you have made your choices from the drop-down box, and pressed Submit, a do-file will then be created: this will be automatically run in Stata and the coefficient estimates and standard errors will be returned for each model you fit* in the yellow table, with a more complete table of results appearing, in sequence for each model, further below.
* NB: up to six model-fitting opportunities are available, but note you don't have to use all of these - fitting four should be sufficient to answer the questions.
Quiz questions
The quiz questions will appear here once they've loaded...
How many quiz questions did you get right?
Once you have submitted your choices, it'll indicate here how many you got correct...
Fitting your models
Table of coefficient estimates (with SE) from all models run
Once the first model has been run, the compiled table of coefficient estimates and standard errors will appear here. As you then run each subsequent model (up to six), the estimates from those will be added too...
| Parameter | MODEL 1, response: | MODEL 2, response: | MODEL 3, response: | MODEL 4, response: | MODEL 5, response: | MODEL 6, response: | ||||||
| cons | () | () | () | () | () | () | ||||||
| sep_2 | () | () | () | () | () | () | ||||||
| sep_3 | () | () | () | () | () | () | ||||||
| alc | () | () | () | () | () | () | ||||||
| bmi | () | () | () | () | () | () | ||||||
| log_ggt | () | () | () | () | () | () | ||||||
| sbp | () | () | () | () | () | () | ||||||
| log_ggt_alc | () | () | () | () | () | () |
Choose your explanatory variables
The list of explanatory variables will appear here; you can then choose which of these you want to include in your model...
Using the box below you can choose which response variable and explanatory variables to include in your linear regression models...
...once you have made your choices, and pressed Submit, a do-file will be created; this will be run in Stata, and the results will be compiled in a table, above, as well as in more detailed results tables returned below.
Remember, you have up to six model-fitting opportunities.
Model 1: 6 model-fitting opportunities remaining...
Model 2: 5 model-fitting opportunities remaining...
Model 3: 4 model-fitting opportunities remaining...
Model 4: 3 model-fitting opportunities remaining...
Model 5: 2 model-fitting opportunities remaining...
Model 6: this is the last model-fitting opportunity...
The specific results table for Model 1 will appear here.
Model 1: Results Table
| Parameter | Coefficient | Standard Error | 95% C.I. lower | 95% C.I. upper | t | p>|t| |
Again, once you have made your choices, the specific results table for Model 2 will appear here.
Model 2: Results Table
| Parameter | Coefficient | Standard Error | 95% C.I. lower | 95% C.I. upper | t | p>|t| |
Again, once you have made your choices, the specific results table for Model 3 will appear here.
Model 3: Results Table
| Parameter | Coefficient | Standard Error | 95% C.I. lower | 95% C.I. upper | t | p>|t| |
Again, once you have made your choices, the specific results table for Model 4 will appear here.
Model 4: Results Table
| Parameter | Coefficient | Standard Error | 95% C.I. lower | 95% C.I. upper | t | p>|t| |
Again, once you have made your choices, the specific results table for Model 5 will appear here.
Model 5: Results Table
| Parameter | Coefficient | Standard Error | 95% C.I. lower | 95% C.I. upper | t | p>|t| |
Again, once you have made your choices, the specific results table for Model 6 will appear here.
Model 6: Results Table
| Parameter | Coefficient | Standard Error | 95% C.I. lower | 95% C.I. upper | t | p>|t| |
Answers
Once you have submitted your answers on the previous page, you'll find out here which questions you got correct / incorrect, and why...