professional/wow-inequalities
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

refactor(code): Split validity calc and matrix extract

Browse source 
Validity calculation belongs to the modelling, so we put it into the
validity module.

Extracting our matrix is a processing step so we made its own matrix
module and put it in their.
Should hopefully provide better separation of concerns going forward.
This commit is contained in:
Marty Oehme 2024-02-16 11:25:19 +01:00
parent 8333bbe9be
commit fac7d4c86a
Signed by: Marty
GPG key ID: EDBF2ED917B2EF6A
5 changed files with 103 additions and 89 deletions
Download patch file Download diff file Expand all files Collapse all files

66
src/model/validity.py
View file

@ -1,9 +1,8 @@
import math
from typing import cast
from pandas import DataFrame
from src import prep_data
def _binned_strength(strength: float) -> str:
if strength < 3.0:
@ -28,15 +27,74 @@ def _combined_validities(
return r"\-"
def calculate(
df: DataFrame, repr_col: str = "representativeness", method_col: str = "method"
) -> DataFrame:
EXT_COL_NAME: str = "external_validity"
INT_COL_NAME: str = "internal_validity"
cols = {EXT_COL_NAME: 0.0, INT_COL_NAME: 0.0}
vd = df[
(df["design"] == "quasi-experimental") | (df["design"] == "experimental")
].copy()
vd.assign(**cols)
vd = cast(DataFrame, vd)
vd[repr_col] = vd[repr_col].fillna("")
vd[method_col] = vd[method_col].fillna("")
# needs to check national before subnational and census, subnational before local
vd.loc[
vd[repr_col].str.contains("|".join(["national", "regional"])), EXT_COL_NAME
] = 4.0
vd.loc[vd[repr_col].str.contains("census"), EXT_COL_NAME] = 5.0
vd.loc[vd[repr_col].str.contains("subnational"), EXT_COL_NAME] = 3.0
vd.loc[vd[repr_col].str.contains("local"), EXT_COL_NAME] = 2.0
# needs to go lowest to highest in case of multiple mentioned approaches
vd.loc[
vd[method_col].str.contains(
"|".join(["OLS", "ordinary.least.square", "logistic.regression"])
),
INT_COL_NAME,
] = 2.0
vd.loc[
vd[method_col].str.contains("|".join(["DM", "discontinuity.matching"])),
INT_COL_NAME,
] = 3.0
vd.loc[
vd[method_col].str.contains(
"|".join(["DID", "difference.in.diff", "diff.in.diff", "triple.diff"])
),
INT_COL_NAME,
] = 3.0
vd.loc[
vd[method_col].str.contains(
"|".join(["PSM", "propensity.score.matching", "score.matching"])
),
INT_COL_NAME,
] = 3.5
vd.loc[
vd[method_col].str.contains("|".join(["IV", "instrumental.variable"])),
INT_COL_NAME,
] = 4.0
vd.loc[
vd[method_col].str.contains("|".join(["RD", "regression.discontinuity"])),
INT_COL_NAME,
] = 4.5
vd.loc[vd[method_col].str.contains("RCT"), INT_COL_NAME] = 5.0
return vd
def add_to_findings(
findings_df: DataFrame, studies_by_intervention: DataFrame
) -> DataFrame:
valid_subset = (
prep_data.calculate_validities(studies_by_intervention)[
calculate(studies_by_intervention)[
["internal_validity", "external_validity", "citation"]
]
.fillna(1.0)
.drop_duplicates(subset=["citation"]) # type: ignore
.drop_duplicates(subset=["citation"]) # type: ignore
.sort_values("internal_validity")
)