CODEWITHSUNDEEP
pythonpandasstatisticspercentagedifference
honeymoon
honeymoon

Reputation: 2520

Percent difference calculation

I want to know whether the following calculation could be simplified or alternative calculations could be used:

id      | group | value
01      |   1   |   5
01      |   1   |   6
02      |   1   |   35
02      |   1   |   40
03      |   1   |   90
03      |   1   |   95
control |   1   |   50
control |   1   |   60
04      |   2   |   35
04      |   2   |   36
05      |   2   |   15
05      |   2   |   10
06      |   2   |   20
06      |   2   |   25
control |   2   |   30
control |   2   |   40

First an average of all controls per group is calculated:

id      | group | value_mean
control |   1   |   55
control |   2   |   35

Then the mean difference relative to the control for each id is calculated:

id      | group | value_mean_percent
01      |   1   |   10
02      |   1   |   68
03      |   1   |   168
04      |   2   |   101
05      |   2   |   36
06      |   2   |   64

Example for id=1 (5+6)/2*100/55 = 10 %

Is there another way to interpret the results of such data?

Upvotes: 1

Views: 92

Answers (1)

StupidWolf
StupidWolf

Reputation: 46888

I think you can define a function that works on a subgroup of the data frame, where it groups value by id, calculates mean and returns everything divided by control:

def func(da):
    da = da.groupby('id')['value'].mean()
    return 100*da[da.index != "control"]/da['control']

Then you just group the data.frame by group, apply this function, and reset index:

import pandas as pd
import numpy as np

df = pd.DataFrame({"id":np.repeat(['01','02','03','control','04','05','06','control'],2),
                  "group":np.repeat([1,2],8),
                   "value":[5,6,35,40,90,95,50,60,35,36,15,10,20,25,30,40]})

df.groupby('group').apply(func).reset_index()

    group   id  value
0   1   01  10.000000
1   1   02  68.181818
2   1   03  168.181818
3   2   04  101.428571
4   2   05  35.714286
5   2   06  64.285714

Upvotes: 1

Related Questions