Some sample solutions are available in the solution script for this practical SB1-CompStats-Third-Prac.R
The data in the file baseball.txt are a subset of data collected on professional American baseball players. Focus on the following variables:
1. salary – The yearly salary of each player (there are 59 NA’s in salary – remove these rows)
2. homeruns_career – The total number of homeruns a player made over their career
3. division – The division a player is in (either W or E)
We havent covered this in lectures yet but it is interesting to look at the code in the answers to see an example of what we will be doing and explore the behavior and check the variable dependence.
You may wish to focus on the next block of questions for this prac.
a.1 Examine the relationship between the log of outcome variable salary and explanatory variables homeruns_career and division using exploratory plots.
a.2 Can we predict salaries with the predictor variable homeruns_career? Perform a regression of response variable log_salary on the variable homeruns_career. Plot the regression lines onto the data and comment on the fit. Give an interpretation of the model, keeping in mind that the salaries are reported on a log-scale. Comment on the effect of extreme observations in homeruns_career.
a.3 Try a smoothing procedure, for example a Kernel smoother ksmooth() or local polynomial regression locpoly(). Plot the fit for various smoothing parameters and discuss what you find. Investigate also lowess() and dpill().
a.4 Could anything be done to improve the fit?
a.5 Optional Extensions: (i) use LOO-CV to estimate the bandwith in a local-polynomial fit of degree 1 (see lecture code L6.R for something nearly identical); (ii) Calculate the smoothing matrix ( in this case (the expression for ( is given in PS2) and verify the fit by hand. Calculate the DOF for GCV, trace„(”, and the DOF for variance estimation, 2trace„(”−trace„()(”.
b.1 Are the salaries similar in the two divisions? Examine whether the distributions of player salaries are the same in both leagues. Use the Wilcoxon test with the normal approximation and calculate by hand the value of the test statistic, its expected value and variance under the null hypothesis of no difference and give the test result.
b.2 The data contain ties. Explain how to simulate the distribution of the test statistic under the null. Implement this procedure and check that the significance levels you computed in the previous question are robust to the effect of ties.
b.3 Test your result by implementing the Wilcoxon test in R and give a point estimate (using the Lehmann-Hodges estimator) and a confidence interval for the difference in median salary between the two leagues. State any assumptions.
c.1 Consider the possibility that walks and runs have the same median. Why should we treat these data as paired?
c.2 Find a confidence interval for the difference in medians and use it to test for equal medians at 95%. What do you conclude? State any assumptions.
Write a short self-contained report introducing the data, clarifying and formalising the question, outlining and justifying your choice of statistical methods (smoother and test) and presenting your results and conclusions in a clear and physically interpretable manner. Draw attention to any assumptions you make in the course of selecting a test and what criterion was used to choose the smoother.
The dataset rainfall.csv available on canvas contains monthly rainfall data for Oxford and Cambridge. This data is from
Total monthly rainfall in mm is given for both locations for the period between January 2014 and December 2020. Perform exploratory data analysis, including at least plotting a smooth of rainfall for each location over months. Perform a test of whether the rainfalls in two locations have the same distributions against a suitable alternative.