example stats

[kingjr]

@qasfb

import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm, ttest_1samp

%matplotlib inline

# Fake data for clarity
n_times = 200
time = np.linspace(-.100, 1., n_times)
word_length = norm.pdf(time, .150, .05)
word_freq = norm.pdf(time, .400, .15) * 3

E = list()
for subject in range(102):
    length = word_length + np.random.randn(n_times)
    freq = word_freq + np.random.randn(n_times)
    plt.plot(time, length, color='C0', lw=.1)
    plt.plot(time, freq, color='C1', lw=.1)
    E.append([length, freq])

E = np.array(E)
plt.plot(time, np.mean(E, 0)[0], color='C0')
plt.plot(time, np.mean(E, 0)[1], color='C1')

# example stat
baseline = 0
t150 = 40
t400 = 60
length, freq = 0, 1

for t in (0., .150, .400):
    t_idx = np.where(time>=t)[0][0]
    mean_length = E[:, length, t_idx].mean(0)
    std_length = E[:, length, t_idx].std(0)

    mean_freq = E[:, freq, t_idx].mean(0)
    std_freq = E[:, freq, t_idx].std(0)

    tval, pval = ttest_1samp(E[:, length, t_idx] - E[:, freq, t_idx], 0)

    print('at %.3f ms, word freq (mean=%.3f +/- %.3f) != word length (mean=%.3f +/- %.3f): t=%.2f p=%.5f' % (
          t, mean_length, std_length, mean_freq, std_freq, tval, pval))