Evalute Matches

Data fusion / statistical matching results are most appropriately used at an aggregate level, as there is no way to guarantee the donated values would have been observed for the individual records. Thus, donated data is usually either analyzed at an aggregate level for inferential statistics or parameter estimates, or used in subsequent modeling where the reality of the individual value is less important. However, its still often desired to have realistic record matches to justify the donated values and increase confidence in insights derived from them.

The easiest way to evaluate individual matches is to consider the common variables (those used in the matching) for the matched pairs. How well these variables align gives an indication of the accuracy of the match. This is esspecially important when using nearest neighbor methods. Agreement then reinforces the assumption that records that “look alike” will “act alike”, justifying the donated variable.

datafusionsm.evaluation.match_accuracy(matches, donors, recipients, feats)

Calculate and display the accuracy of the expected matching variables post-fusion.

Parameters

• matches (array-like) – list of matching id pairs resulting from fusion

• donors (pandas.DataFrame) – donating records

• recipients (pandas.DataFrame) – receiving records

• feats (array-like) – what features were fused on and should be evaluated

Returns

• str – formatted output string containing feature evaluation

• results (pandas.DataFrame) – detailed counts of feature matches