konosp/propensity-score-matching

propensity-score-matching's Issues

About your knn radius

In your jupyter notebook, you use Nearerst Neighbors to identify matching candidates.

caliper = np.std(df_data.propensity_score) * 0.25 knn = NearestNeighbors(n_neighbors=10 , p = 2, radius=caliper)

In this line, why you choose caliper as radius? Thank you.

I'm getting a key-error when running the "perfom_matching_v2"-function

I want to perform propensity score matching using your jupyter notebook template. I have a sample of treated and untreated individuals where the treated group is smaller compared to the untreated group. Moreover, my outcome is continuous, and not categorical compared to yours - which should be fine aswell. I can run your code until the "perfom_matching_v2"-function (see below; cell 21 in your code). When I run it, I get KeyError: 37. Any idea why this could happen and how to solve it?

def perfom_matching_v2(row, indexes, df_data):
    current_index = int(row['index']) # Obtain value from index-named column, not the actual DF index.
    prop_score_logit = row['propensity_score_logit']
    for idx in indexes[current_index,:]:
        if (current_index != idx) and (row.treatment == 1) and (df_data.loc[idx].treatment == 0):
            return int(idx)
         
df_data['matched_element'] = df_data.reset_index().apply(perfom_matching_v2, axis = 1, args = (indexes, df_data))