您可以使用shift()和np.where()完成此操作

df['Change in Mental Health (TP1 and 3)'] = df['Mental Health Score'] - df['Mental Health Score'].shift(2)
df['Change in Mental Health (TP1 and 3)'] = np.where(df['Timepoint'] != 3, 0, df['Change in Mental Health (TP1 and 3)']).astype(int)
df

用groupby和where试试：

#sort by Timepoint if needed
#df = df.sort_values("Timepoint")

changes = df.groupby("User")["Mental Health Score"].transform('last')-df.groupby("User")["Mental Health Score"].transform('first')
df["Change"] = changes.where(df["Timepoint"].eq(3))

>>> df
   User  Timepoint  Mental Health Score  Change
0  Bill          1                    5     NaN
1  Bill          2                   10     NaN
2  Bill          3                   15    10.0
3   Wiz          1                   10     NaN
4   Wiz          2                   10     NaN
5   Wiz          3                   15     5.0
6   Sam          1                    5     NaN
7   Sam          2                    5     NaN
8   Sam          3                    5     0.0

正如在注解中已经指出的，您可以在User上groupby您的 Dataframe ，并在Mental Health Score上计算差异
我在这里放了一段代码来演示

def _overall_change(scores):
  return scores.iloc[-1] - scores.iloc[0]

person = df.groupby('User')['Score'].agg(_overall_change)

使用groupby和merge：

g = df.sort_values(by='Timepoint').groupby('User')['Mental Health Score']

s = pd.concat({3: g.last()-g.first()})
#    User
# 3  Bill    10
#    Sam      0
#    Wiz      5
# Name: Mental Health Score, dtype: int64

df.merge(s, left_on=['Timepoint', 'User'], right_index=True, how='left')

输出：

User  Timepoint  Mental Health Score_x  Mental Health Score_y
0  Bill          1                      5                    NaN
1  Bill          2                     10                    NaN
2  Bill          3                     15                   10.0
3   Wiz          1                     10                    NaN
4   Wiz          2                     10                    NaN
5   Wiz          3                     15                    5.0
6   Sam          1                      5                    NaN
7   Sam          2                      5                    NaN
8   Sam          3                      5                    0.0

这是另一个可能的解决方案：

import pandas as pd

def calculate_change(mhs):
    mhs = list(mhs)
    return mhs[-1] - mhs[0]

df = df.sort_values(["User", "Timepoint"])
diff = df.groupby('User')['Mental Health Score'].agg(calculate_change)

df = pd.merge(df, diff, how='left', left_on='User', right_index=True)
df.columns = ['User', 'Timepoint', 'Mental Health Score', 'Change']
df['Change'] = df['Change'].loc[df['Timepoint']==3]
print(df)

输出

User  Timepoint  Mental Health Score  Change
0  Bill          1                    5     NaN
1  Bill          2                   10     NaN
2  Bill          3                   15    10.0
3   Wiz          1                   10     NaN
4   Wiz          2                   10     NaN
5   Wiz          3                   15     5.0
6   Sam          1                    5     NaN
7   Sam          2                    5     NaN
8   Sam          3                    5     0.0