NBA Trends

[oldoc63]

• You'll analyze data from the NBA (National Basketball Association) and explore possible associations.
• We limited the data to just 5 teams and 10 columns (plus one constructed column, point_diff, the different between pts and opp_pts).
• We will create several charts and tablets in this project, so you'll need to use plt.clf() between plots in your code so that the plots don't layer on top of one another.

[oldoc63]

Analyzing relationships between Quant and Categorical

• In script.py the data has been subsetted for you into two smaller datasets: games from 2010 (named nba_2010) and games from 2014 (named nba_2014).
• Suppose you want to compare the knicks to the nets with respect to points earned per game. Using the pts column from the nba_2010 DataFrame, create two series named knicks_pts_10 (fran_id = "Knicks") and net_pts_10 (fran_id = "Nets") that represent the points each team has scored in their games.

[oldoc63]

• Calculate the difference between the two teams’ average points scored and save the result as diff_means_2010. Based on this value, do you think fran_id and pts are associated? Why or why not?

[oldoc63]

• Rather than comparing means, it's useful look at the full distribution of values to understand whether a difference in means is meaningful. Create a set of overlapping histograms that can be used to compare the points scored for the Knicks compared to the Nets. Use the series you created in the previous step and the code below to create the plot. Do the distributions appear to be the same?

• Repeat the same steps for the 2014 season.

[oldoc63]

• Generate side by side boxplots with points scored (pts) on the y axis and team (fran_id) on the x axis.

[oldoc63]

Analyzing relationships between categorical variables

• We'd like to know if teams tend to win more games at home compared to away.
• The variable game result indicates whether a team won a particular game ('W', 'L'). The variable game_location indicates whether a team was playing at home or away ('H', 'A').
• Data scientist will often calculate a contingency table of frequencies to help them determine if categorical variables are associated.

[oldoc63]

• You can convert this table of frequencies to a table of proportions by dividing the values in location_result_frequency by the total of observations

[oldoc63]

• Using the contingency table we created in Task 6 (counts not proportions), calculate the expected contingency table (if there were no association) and the Chi-Square statistic.

[oldoc63]

Analyzing relationships between quantitative variables

• For each game, 538 has calculated the probability that each team will win the game. We want to know if teams with a higher probability of winning also tend to win games by more points.
• In the data, 538's prediction is saved as forecast. The point_diff column gives the margin of victory/defeat for each team (positive values mean that the team won; negative values mean that they lost).
• Using nba_2010, calculate the covariance between forecast (538's projected win probability) and point_diff (the margin of victory/defeat) in the dataset. Call this point_diff_forecast_cov. Save and print your result.
• Use the np.cov() function to calculate the covariance. Pass the dataframe columns, forecast and point_diff as arguments to np.cov(). You can identify the covariance between two by finding the number that is represented twice in the matrix.

[oldoc63]

• Because 538's forecast is reported as a probability (not a binary), we can calculate the strength of the correlation. Using nab_2010, calculate the correlation between forecast and point_diff. Call this point_diff_forecast_corr.
• Use personr from the scipy.stats package to calculate correlation.
• Generate a scatter plot of forecast and point_diff