How to Do One-way ANOVA Using Python (1) | F Test | Analysis Of ...

How to do one-way ANOVA using Python Originally posted by Python Psychologist

What is repeated measures ANOVA? ●

●

A repeated-mea repeated-measures sures ANOVA (rmANOVA) is extending the analysis of variance tosituations using repeated-measures repeated-measures research designs. (e.g., in which all subjects have been through each condition)

Logic of rmANOVA and independent measures ANOVA is similar; many formulas are, basically the same, ○ ○

Second stage of analysis in rmANOVA to get the individual differences subs the error term.

What is repeated measures ANOVA? ●

 A repeated-me repeated-measures asures design eliminates individual i ndividual differences from the between-treatments variability because the same subjects go through each treatment condition.

●

The F-ratio needs to be balanced with the calculation such that the individual differences are eliminated from the F-ratio.

●

In the end we get a similar test statistic as in an ordinary ANOVA but all individual differences are removed. Thus, there are no individual differences between treatments.

What is repeated measures ANOVA? ●

The variability due to individual differences is not a component of the numerator of the F-ratio.

●

Individual differences must also be removed from the denominator of the F ratio to maintain a balanced ratio with an expected value of 1.00 when there is no treatment effect:

What is repeated measures ANOVA? This can be accomplished by two stages. Note, SS stands for Sum of Squares.

1.

First, Firs t, th the e tot total al va vari riab abil ilit ityy (SS (SS total) total) is partitioned into variability between-treatments between-treatments (SS ( SS between) between) and within-treatments (SS ( SS within). within). Individual differences do not appear in SS between due to that the same sample of subjects were measured in every treatment. Individual differences do play a role in SS in SS total because total because the sample contains different subjects.

2.

Second, we measure Second, measure the individ individual ual differ difference encess by calculat calculating ing the the variabili variability ty between between subje subjects, cts, or SS subjects. SS value is subtracted from SS within and we obtain obtain variability due to sampling error, SS error,  SS erro

Doing one-way ANOVA in Python

import pandas as pd  import numpy as np

In the code to the left we import the needed python librares.

from scipy import stats def calc_grandmean(data, columns):   "“” Takes a pandas dataframe and calculates the grand mean data = dataframe columns = list of column names names with the response variables "“”  gm = np.mean(data[colum np.mean(data[columns].mean()) ns].mean()) return gm

I also created a function to calculate the grand mean.

Doing one-way ANOVA in Python ##For createing example data  X1 = [6,4,5,1,0,2]

I then create some data using 3 lists and Pandas DataFrame.

 X2 = [8,5,5,2,1,3]  X3 = [10,6,5,3,2,4] df = pd.DataFrame({‘ pd.DataFrame({‘Subid’:xrange(1 Subid’:xrange(1,, len(X1)+1), “X1”:X1, “X2”:X2,

After data creation we calculate the grand mean, subject mean, and column means.

“X3”:X3}) #Grand mean grand_mean = calc_grandmean(df, ['X1’, 'X2’, 'X3’]) df['Submean’] df['Submea n’] = df[['X1’, 'X2’, 'X3’]].mean(axis=1) column_means = df[['X1’, 'X2’, 'X3’]].mean(ax 'X3’]].mean(axis=0) is=0)

All means are, later, going to be used in the ANOVA calculation.

Doing one-way ANOVA in Python n = len(df['Subid’] len(df['Subid’])) k = len(['X1’, 'X2’, 'X3’])

We now go on to get the sample size and the number of levels of the within-subject factor.

#Degree of Freedom ncells = df[['X1’,'X2 df[['X1’,'X2’,'X3’]].size ’,'X3’]].size dftotal = ncells - 1

After this is done we need to calculate the degree of freedoms.

dfbw = 3 - 1 dfsbj = len(df['Subid’]) - 1 dfw = dftotal - dfbw  dferror = dfw - dfsbj

All of these are going to be used in the calculation of sum of squares and means square, and finally the F-ratio.

Doing one-way ANOVA in Python ●

●

Sum of Squares Between is calculated using this formula:

Python code: SSbetween = sum(n*[(m - grand_mean)**2 for m in column_means])

Doing one-way ANOVA in Python ●

●

Sum of Squares Within is calculated using this formula:

Python code: SSwithin = sum(sum([(df[col] - column_means[i])**2 for i, col in enumerate(df[['X1’, 'X2’, 'X3’]])]))

Doing one-way ANOVA in Python ●

●

Sum of Squares Subjects is calculated using this formula:

Python code: SSsubject = sum(k*[(m -grand_mean)**2 for m in df['Submean’]])

Doing one-way ANOVA in Python ●

●

Sum of Squares Error is calculated using this formula:

Python code: SSerror = SSwithin - SSsubject

Doing one-way ANOVA in Python ●

● ●

We can also calculate the SS total (i.e., The sum of squared deviations of all observations from the grand mean):

Python code: SStotal = SSbetween + SSwithin Although it is not entirely necessary...

Doing one-way ANOVA in Python ●

After we have calculated the Mean square error and Mean square between we can obtain the F-statitistica:

msbetween = SSbetween/dfbetween mserror = SSerror/dferror F = msbetween/mserror

Doing one-way ANOVA in Python ●

By using SciPy we can obtain a p-value. We start by setting our alpha to .05 and then we get our p-value. alpha = 0.05 p_value = stats.f.sf(F, 2, dferror)

●

That was it! If you have any question please let me know.

●

I blog images related to data, Python, statistics, and psychology related stuff on my tumblr: http://pythonpsychologist.tumblr.com/