Milestone 2 Feedback

[KateJohnson]

Survey:

• Be careful about collecting IP address, although I note that you used a UBC hosted platform to comply with privacy policy

EDA

• Good summary of the data and description.
• In general I really liked your visualizations, however I thought they lacked focus. They should focus on the relationship between the dependent variable (sustainability attitude) and independent variable of interest (recycling frequency). Then stratify by levels of your other variables to identify confounding.
• In the comparison of sustainability ratings before and after the video, it doesn’t look like you filtered the before data to only those people who answered the after question (minor point).
• I really liked the scatter plots to assess whether recycling or age affects sustainability rating change, but again you’re not really getting at confounding here.

Next steps:

• Do you feel that you got enough response to assess sustainability attitude change? If so, do you think an average shift of ~1 point is a meaningful change? I might suggest focusing on initial sustainability rating as the main outcome, and assessing the impact of recycling frequency while adjusting for confounding.

[heathervant]

Thanks for the feedback @KateJohnson ! We will definitely remove the ip addresses. We were too excited about exploring all of our variable I guess. We will keep the EDA more focused for milestone 3. After some further thought, we would like to make our question: Does one's attitudes towards the importance of sustainability affect how often they recycle? ie have attitude be the continuous explanatory X variable, and the y be our categorical response variable. Just clarifying as I think this was mixed up in our milestone 2 introduction. (apologies). But if we set it up this way, we can do linear regression if we assume that the frequency or recycling categories (5) are equally spaced (which I think is a fair assumption) and we also talked to Vincenzo and would like to try doing an ordinal regression as well, but first grouping the opinions into 3 groups so that there are less categories to worry about. What do you think?

[KateJohnson]

Hi @heathervant, I really like that plan, and both analysis approaches you suggest. Just be sure to check the linear regression assumptions- look at a residual plot, and in particular a q-q plot to see if they're normally distributed. If it the fit is so-so, do an ordinal analysis (check the proportional odds assumption), and compare the coefficients. If they're similar, then you're probably fine with either method. In marking this, I'll probably be focusing on whether you tested modelling assumptions, and whether your interpretation of the findings is appropriate. Let me know if I can help with anything else!