lakillo
commented
1 year ago - Representativeness of the data should be clearly defined e.g. missing data (what data do we have at our disposal? Is there any data we need that we don't have? (mental health, unemployed people). Bias / skewness
- Effect of cultural / social / political contexts of the data for some countries
- Important issue to discuss with the domain expert (postdoc)
- Using biased data but wanting to influence policy
- Storage after the analysis (where, how long)
- Maintenance of any public-facing output
- Tools and data used need to be transparent and reproducible for potential end use in policy (e.g. if data for other years becomes available)
- Some variables not represented for some countries
- Self-reporting nature of data is affected by cultural context (e.g. some cultures may report more openly)
- Better ways to measure health than reported here? (life expectancy etc)
- Limitations important to report in final publication
- Making sure everything is licensed appropriately (data, model framework)
- Understanding the purpose of the survey (e.g. who's funding the research, who's asking the questions)
- Could the outcomes of the project be used to oppress certain groups? (e.g. health insurance)
- Gender is reported in a restricted binary
- Check for biases within the group and blind spots
- Highlight the limitations of model and data, especially when presenting, to combat myths about data science