Red and White Wine - Data Science Primer

Is white wine more acidic due to fermentation without skins? Hypothesis testing example: Difference of means

import numpy as np
from datascience import *
import matplotlib.pyplot as plt
%matplotlib inline

file = 'winequality_redwhite.csv'

wine = Table.read_table(file)
wine

wine.group('type')

Hypothesis¶

White wines are more acidic than red

Why: White wines are usually fermented without skins, emphasizing acidity, while red wines are fermented with skins and seeds, imparting tannins that can mask the perception of acidity. Lodi Growers

Null hypothesis¶

Difference in pH between individaul red and white wines is random

wine_group = wine.group('type',np.mean)
wine_group

Difference of means¶

use index [1] and [0] to access elements of array created from column

pHdiff = wine_group.column('pH mean')[1] - wine_group.column('pH mean')[0]
pHdiff

-0.12284655630267016

White has lower pH which means more acidic

Simulate Null Distribution¶

pHdiff_sim = []

for i in np.arange(300):
    wine_group_s = wine.sample().group('type',np.mean)
    sim_diff = wine_group_s.column('pH mean')[1] - wine_group_s.column('pH mean')[0]
    pHdiff_sim.append(sim_diff)

len(pHdiff_sim)

300

plt.hist(pHdiff_sim)
plt.axvline(pHdiff)

p = np.count_nonzero( np.array(pHdiff_sim) >= 0 )/len(pHdiff_sim)
p

0.0

p < 0.05 Data suports Hypothesis, discard Null hypothesis¶

Alternate Simulation of Null Distribution¶

pHdiff_sim = []
winetype = wine.column('type')

for i in np.arange(300):
    winetype_s = np.random.shuffle(winetype)
    wine_s = wine.with_column('Stype',winetype_s)    
    wine_group_s = wine_s.group('type',np.mean)
    sim_diff = wine_group_s.column('pH mean')[1] - wine_group_s.column('pH mean')[0]
    pHdiff_sim.append(sim_diff)

plt.hist(pHdiff_sim)

(array([  2.,   5.,   9.,  17.,  45.,  68.,  66.,  49.,  25.,  14.]),
 array([-0.01543567, -0.01281095, -0.01018624, -0.00756152, -0.00493681,
        -0.00231209,  0.00031262,  0.00293734,  0.00556205,  0.00818677,
         0.01081148]),
 <BarContainer object of 10 artists>)

plt.hist(pHdiff_sim)
plt.axvline(pHdiff)

p = np.count_nonzero( pHdiff_sim <= pHdiff )/len(pHdiff_sim)
p

0.0