Analysis of results of the 2015 FINA World Swimming Championships
In this chapter, you will practice your EDA, parameter estimation, and hypothesis testing skills on the results of the 2015 FINA World Swimming Championships. This is the Summary of lecture "Case Studies in Statistical Thinking", via datacamp.
- Introduction to swimming data
- Do swimmers go faster in the finals?
- How does the performance of swimmers decline over long events?
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import dc_stat_think as dcst
plt.rcParams['figure.figsize'] = (10, 5)
Introduction to swimming data
- Strokes at the World Championships
- Freestyle
- Breaststroke
- Butterfly
- Backstroke
- Events at the World Championships
- Defined by gender, distance, stroke
- Rounds of events
- Heats: First round
- Semifinals: Penultimate round in some events
- Finals: The final round; the winner is champion
- Data source
- Data are freely available from OMEGA
- Domain-specific knowledge
- Imperative
- An absolute pleasure
swim = pd.read_csv('./dataset/2015_FINA.csv', skiprows=4)
swim.head()
mens_200_free_heats_df = swim[(swim['gender'] == 'M') &
(swim['distance'] == 200) &
(swim['stroke'] == 'FREE') &
(swim['round'] == 'PRE') &
(swim['split'] == 4)]
mens_200_free_heats = mens_200_free_heats_df['cumswimtime'].unique()
x, y = dcst.ecdf(mens_200_free_heats)
# Plot the ECDF as dots
_ = plt.plot(x, y, marker='.', linestyle='none')
# Label axes and show plot
_ = plt.xlabel('time (s)')
_ = plt.ylabel('ECDF')
We see that fast swimmers are below 115 seconds, with a smattering of slow swimmers past that, including one very slow swimmer.
200 m free time with confidence interval
Now, you will practice parameter estimation and computation of confidence intervals by computing the mean and median swim time for the men's 200 freestyle heats. The median is useful because it is immune to heavy tails in the distribution of swim times, such as the slow swimmers in the heats.
mean_time = np.mean(mens_200_free_heats)
median_time = np.median(mens_200_free_heats)
# Draw 10,000 bootstrap replicates of the mean and median
bs_reps_mean = dcst.draw_bs_reps(mens_200_free_heats, np.mean, size=10000)
bs_reps_median = dcst.draw_bs_reps(mens_200_free_heats, np.median, size=10000)
# Compute the 95% confidence intervals
conf_int_mean = np.percentile(bs_reps_mean, [2.5, 97.5])
conf_int_median = np.percentile(bs_reps_median, [2.5, 97.5])
# Print the result to the screen
print("""
mean time: {0:.2f} sec.
95% conf int of mean: [{1:.2f}, {2:.2f}] sec.
median time: {3:.2f} sec.
95% conf int of median: [{4:.2f}, {5:.2f}] sec.
""".format(mean_time, *conf_int_mean, median_time, *conf_int_median))
Do swimmers go faster in the finals?
-
Question : Do swimmers swim faster in the finals than in other rounds?
- Individual swimmers, or the whole field?
- Faster than heats? Faster than semifinals?
- For what strokes? For what distances?
-
Question: Do individual female swimmers swim faster in the finals compared to the semifinals?
- Events: 50, 100, 200 meter freestyle, breaststroke, butterfly, backstroke
-
Fractional improvement
$$ f = \frac{\text{semifinals time} - \text{finals time}}{\text{semifinals time}} $$
- Sharpened questions
- What is the frational improvement of individual female swimmers from the semifinals to the finals?
- Is the observed fractional improvement commensurate with there being no difference in performance in the semifinals and finals?
EDA: finals versus semifinals
First, you will get an understanding of how athletes' performance changes from the semifinals to the finals by computing the fractional improvement from the semifinals to finals and plotting an ECDF of all of these values.
The arrays final_times and semi_times contain the swim times of the respective rounds. The arrays are aligned such that final_times[i] and semi_times[i] are for the same swimmer/event. If you are interested in the strokes/events, you can check out the data frame df in your namespace, which has more detailed information, but is not used in the analysis.
women_swim_df = swim[(swim['gender'] == "F") &
(swim['stroke'] != "MEDLEY") &
(swim['distance'].isin([100, 50, 200])) &
(swim['round'].isin(['SEM', 'FIN'])) &
(swim['splitdistance'] == swim['distance'])]
women_swim_df.head(n = 5)
women_swim_df = women_swim_df[['athleteid', 'stroke', 'distance', 'lastname', 'cumswimtime', 'round']]
women_swim_df_fin = women_swim_df.loc[(women_swim_df['round'] == 'FIN')]
women_swim_df_sem = women_swim_df.loc[(women_swim_df['round'] == 'SEM')]
women_swim_df_w = women_swim_df_fin.merge(women_swim_df_sem, how = 'left', on = ['athleteid', 'stroke', 'distance', 'lastname'])
df = women_swim_df_w.rename(index = str, columns = {"cumswimtime_x" : "final_swimtime", "cumswimtime_y" : "semi_swimtime"})
df = df[['athleteid', 'stroke', 'distance', 'lastname', 'final_swimtime', 'semi_swimtime']]
final_times = df['final_swimtime'].values
semi_times = df['semi_swimtime'].values
df
f = (df['semi_swimtime'] - df['final_swimtime']) / df['semi_swimtime']
# Generate x and y values for the ECDF: x, y
x, y = dcst.ecdf(f)
# Make a plot of the ECDF
_ = plt.plot(x, y, marker='.', linestyle='none')
# Label axes
_ = plt.xlabel('f')
_ = plt.ylabel('ECDF')
The median of the ECDF is juuuust above zero. But at first glance, it does not look like there is much of any difference between semifinals and finals.
f_mean = np.mean(f)
# Get bootstrap reps of means: bs_reps
bs_reps = dcst.draw_bs_reps(f, np.mean, size=10000)
# Compute confidence intervals: conf_int
conf_int = np.percentile(bs_reps, [2.5, 97.5])
# Report
print("""
mean frac. diff.: {0:.5f}
95% conf int of mean frac. diff.: [{1:.5f}, {2:.5f}]""".format(f_mean, *conf_int))
It looks like the mean finals time is juuuust faster than the mean semifinal time, and they very well may be the same. We'll test this hypothesis next.
How to do the permutation test
Based on our EDA and parameter estimates, it is tough to discern improvement from the semifinals to finals. In the next exercise, you will test the hypothesis that there is no difference in performance between the semifinals and finals. A permutation test is fitting for this. We'll get test statistics with following strategy:
- Take an array of semifinal times and an array of final times for each swimmer for each stroke/distance pair.
- Go through each array, and for each index, swap the entry in the respective final and semifinal array with a 50% probability.
- Use the resulting final and semifinal arrays to compute f and then the mean of f.
Generating permutation samples
As you worked out in the last exercise, we need to generate a permutation sample by randomly swapping corresponding entries in the semi_times and final_times array. Write a function with signature swap_random(a, b) that returns arrays where random indices have the entries in a and b swapped.
def swap_random(a, b):
"""Randomly swap entries in two arrays"""
# Indices to swap
swap_inds = np.random.random(size=len(a)) < 0.5
# Make copies of arrays a and b for output
a_out = np.copy(a)
b_out = np.copy(b)
# Swap values
a_out[swap_inds] = b[swap_inds]
b_out[swap_inds] = a[swap_inds]
return a_out, b_out
Hypothesis test: Do women swim the same way in semis and finals?
Test the hypothesis that performance in the finals and semifinals are identical using the mean of the fractional improvement as your test statistic. The test statistic under the null hypothesis is considered to be at least as extreme as what was observed if it is greater than or equal to f_mean.
perm_reps = np.empty(1000)
for i in range(1000):
# Generate a permutation sample
semi_perm, final_perm = swap_random(semi_times, final_times)
# Compute f from the permutation sample
f = (semi_perm - final_perm) / semi_perm
# Compute and store permutation replicate
perm_reps[i] = np.mean(f)
# Compute and print p-value
print('p =', np.sum(perm_reps >= f_mean) / 1000)
The p-value is large, about 0.28, which suggests that the results of the 2015 World Championships are consistent with there being no difference in performance between the finals and semifinals.
How does the performance of swimmers decline over long events?
- Swimming background
- Split: The time is takes to swim one length of the pool
- Quantifying slowdown
- Use women's 800m freestyle heats
- Omit first and last 100 meters
- Compute mean split time for each split number
- Perform linear regression to get slowdown per split
- Perform hypothesis test: can the slowdown be explained by random variation?
- Hypothesis tests for correlation
- Posit null hypothesis: split time and split number are completely uncorrelated
- Simulate data assuming null hypothesis is true
- Use pearson correlation, $\rho$ as test statistic
- Compute p-value as the fraction of replicates that have $\rho$ at least as large as observed
EDA: Plot all your data
To get a graphical overview of a data set, it is often useful to plot all of your data. In this exercise, plot all of the splits for all female swimmers in the 800 meter heats. The data are available in a Numpy arrays split_number and splits. The arrays are organized such that splits[i,j] is the split time for swimmer i for split_number[j].
free_800_w = swim.loc[(swim['gender'] == 'F') &
(swim['stroke'] == 'FREE') &
(swim['distance'] == 800) &
(swim['round'].isin(['PRE'])) &
(~swim['split'].isin([1,2,15,16]))]
free_800_w = free_800_w[['split', 'splitswimtime']]
splits = np.reshape(free_800_w['splitswimtime'].values, (-1, 12))
split_number = free_800_w['split'].unique()
for splitset in splits:
_ = plt.plot(split_number, splitset, linewidth=1, color='lightgray')
# Compute the mean split times
mean_splits = np.mean(splits, axis=0)
# Plot the mean split time
_ = plt.plot(split_number, mean_splits, marker='.', linewidth=3, markersize=12)
# Label axes
_ = plt.xlabel('split number')
_ = plt.ylabel('split time (s)')
You can see that there is wide variability in the splits among the swimmers, and what appears to be a slight trend toward slower split times.
Linear regression of average split time
We will assume that the swimmers slow down in a linear fashion over the course of the 800 m event. The slowdown per split is then the slope of the mean split time versus split number plot. Perform a linear regression to estimate the slowdown per split and compute a pairs bootstrap 95% confidence interval on the slowdown. Also show a plot of the best fit line.
slowdown, split_3 = np.polyfit(split_number, mean_splits, 1)
# Compute pairs bootstrap
bs_reps, _ = dcst.draw_bs_pairs_linreg(split_number, mean_splits, size=10000)
# Compute confidence interval
conf_int = np.percentile(bs_reps, [2.5, 97.5])
# Plot the data with regressions line
_ = plt.plot(split_number, mean_splits, marker='.', linestyle='none')
_ = plt.plot(split_number, slowdown * split_number + split_3, '-')
# Label axes and show plot
_ = plt.xlabel('split number')
_ = plt.ylabel('split time (s)')
# Print the slowdown per split
print("""
mean slowdown: {0:.3f} sec./split
95% conf int of mean slowdown: [{1:.3f}, {2:.3f}] sec./split""".format(
slowdown, *conf_int))
There is a small (about 6 hundreths of a second), but discernible, slowdown per split. We'll do a hypothesis test next.
rho = dcst.pearson_r(split_number, mean_splits)
# Initialize permutation reps
perm_reps_rho = np.empty(10000)
# Make permutation reps
for i in range(10000):
# Scramble_split number array
scrambled_split_number = np.random.permutation(split_number)
# Compute the Pearson correlation coefficient
perm_reps_rho[i] = dcst.pearson_r(scrambled_split_number, mean_splits)
# Compute and print p-value
p_val = np.sum(perm_reps_rho >= rho) / 10000
print('p =', p_val)
The tiny effect is very real! With 10,000 replicates, we never got a correlation as big as observed under the hypothesis that the swimmers do not change speed as the race progresses. In fact, I did the test with a million replicates, and still never got a single replicate as big as the observed Pearson correlation coefficient.