Notes on the individual coursework

The premise is that we want to do plots and calculations to learn something about the data that we have been given.

We are given an outcome e.g. life expectancy and we are asked to justify which variabel gives us the best way to explain life expectancy. We could calculate correlation to use as justification.

We could use root mean square error to see which variable produces the smallest uncertainty. How much uncertainty do we have in out outcome in the first place - what variable lead to the biggest reduction in uncertainty.

Squaring correlation values is another method to justify which variable would be best. r² gives us a standardised scale - the maximum value we could get is 1 but yet our value is 0.175. The root mean square error (root MSE) has units which r² does not.

You can also comment on whether the variables are useful in the first place.

We can also ask Stata to produce confidence bands on our regression graphs. We can either produce a confidence band on where our regression line could be. There’s an error in our y-intercept value as well as our gradient value. We can also produce a confidence band on how far away the real plots are from our line. We could be 95% confident that every real plot would be in a band around the regression line.

First part of the project: we are providing 3 analyses of the data. Tip: produce a scatterplot matrix to guide you on where to analyse.

Add as much details to the graph as possible - make the points slightly transparent, add a regression line etc

When commenting on graph, make more of a discussion than as X goes up, Y goes down. Example: People of working age goes up when population density goes up. People who are retired doesn’t tend to be in high populated areas. Ask why we are seeing a certain correlation.