def tax(flat_tax, total_type="person"):
"""Calculate all metrics given a flat tax.
Args:
flat_tax: Percentage tax rate (0-100).
total_type: Whether to use total population and current tax liability
from SPM units or persons. Either "person" or "spmu".
Defaults to "person".
"""
flat_tax /= 100
spmu["new_tax"] = spmu.spmu_agi * flat_tax
new_revenue = mdf.weighted_sum(spmu, "new_tax", "spm_weight")
change_revenue = new_revenue - totals.loc[total_type].fica_fedtax_ac
ubi = change_revenue / totals.loc[total_type].person
spmu["new_spm_resources"] = (spmu.spm_resources_before_tax +
ubi * spmu.spmu_total_people - spmu.new_tax)
# Merge back to each person.
target_persons = person.merge(spmu[SPM_COLS + ["new_spm_resources"]],
on=SPM_COLS)
target_persons["new_spm_resources_pp"] = (
target_persons.new_spm_resources / target_persons.spmu_total_people)
# Calculate poverty rate
target_persons["new_poor"] = (target_persons.new_spm_resources <
target_persons.spm_povthreshold)
poverty_rate = mdf.weighted_mean(target_persons, "new_poor", "marsupwt")
change_poverty_rate = chg(poverty_rate, initial_poverty_rate)
# Calculate poverty gap
poverty_gaps = np.maximum(spmu.spm_povthreshold - spmu.new_spm_resources,
0)
poverty_gap = (poverty_gaps * spmu.spm_weight).sum()
change_poverty_gap = chg(poverty_gap, initial_poverty_gap)
# Calculate Gini
gini = mdf.gini(target_persons, "new_spm_resources_pp", w="marsupwt")
change_gini = chg(gini, initial_gini)
# Percent winners
target_persons["better_off"] = (target_persons.new_spm_resources >
target_persons.spm_resources)
percent_better_off = mdf.weighted_mean(target_persons, "better_off",
"marsupwt")
return pd.Series({
"poverty_rate": poverty_rate,
"poverty_gap": poverty_gap,
"gini": gini,
"percent_better_off": percent_better_off,
"change_poverty_rate": change_poverty_rate,
"change_poverty_gap": change_poverty_gap,
"change_gini": change_gini,
"change_revenue": change_revenue,
"ubi": ubi,
})
def pv_rate(column):
return (
mdf.weighted_mean(
target_persons[target_persons[column]], "poor", "asecwt"
)
* 100
)
def pov(race, monthly_ubi):
# Total cost and associated tax rate.
cost = monthly_ubi * total_population * 12
tax_rate = cost / total_agi
# Calculate new tax, UBI and resources per SPM unit.
spm["new_spm_resources"] = (
spm.spm_resources - (tax_rate * spm.spm_agi_pos) + # New tax
(12 * monthly_ubi * spm.spm_numper)) # UBI
# Merge back to person.
person2 = person.merge(spm[["spm_id", "new_spm_resources"]], on="spm_id")
# Based on new resources, calculate
person2["new_poor"] = person2.new_spm_resources < person2.spm_povthreshold
# Calculate poverty rate for specified race.
poverty_rate = mdf.weighted_mean(person2[person2[race.lower()]],
"new_poor", "weight")
# Calculate poverty gap for specified race.
poverty_gap = pov_gap(spm[spm["spm_" + race.lower()] > 0],
"new_spm_resources", "spm_povthreshold",
"spm_weight")
poverty_gap_per_capita = (poverty_gap / pop_in_race_spmu[race])
return pd.Series({
"poverty_rate": poverty_rate,
"poverty_gap_per_capita": poverty_gap_per_capita
})
def pov_rate(column):
return (
mdf.weighted_mean(
target_persons[target_persons[column]],
"original_poor",
"asecwt",
)
* 100
)
def test_weighted_mean():
# Test umweighted.
assert mdf.weighted_mean(df, "x") == 8 / 3
# Test weighted.
assert mdf.weighted_mean(df, "x", "w") == 11 / 6
# Test weighted with multiple columns.
assert mdf.weighted_mean(df, ["x", "y"], "w").tolist() == [11 / 6, -3 / 6]
# Test grouped.
mdf.weighted_mean(dfg, "x", "w", "g")
mdf.weighted_mean(dfg, ["x", "y"], "w", "g")
def pov(data, group):
return pd.DataFrame(
mdf.weighted_mean(data, "poverty_flag", "asecwt", groupby=group))
# Calculate totals at both person and SPM unit levels so we can compare and
# calculate poverty gaps.
person_totals = mdf.weighted_sum(person, ["fica_fedtax_ac", "person"],
"marsupwt")
spmu_totals = mdf.weighted_sum(spmu,
["spmu_fica_fedtax_ac", "spmu_total_people"],
"spm_weight")
spmu_totals.index = person_totals.index
totals = pd.concat([person_totals, spmu_totals], axis=1).T
totals.index = ["person", "spmu"]
# Calculate status quo
person["poor"] = person.spm_resources < person.spm_povthreshold
initial_poverty_rate = mdf.weighted_mean(person, "poor", "marsupwt")
spmu["initial_poverty_gap"] = np.maximum(
spmu.spm_povthreshold - spmu.spm_resources, 0)
initial_poverty_gap = (spmu.initial_poverty_gap * spmu.spm_weight).sum()
person["spm_resources_pp"] = person.spm_resources / person.spmu_total_people
initial_gini = mdf.gini(person, "spm_resources_pp", w="marsupwt")
def chg(new, base):
return (100 * (new - base) / base).round(1)
def tax(flat_tax, total_type="person"):
"""Calculate all metrics given a flat tax.
mdf.add_weighted_quantiles(temp, 'e00200', 's006')
cps2 = pd.concat([cps2, temp], ignore_index=True)
cps2['income_bin'] = pd.cut(cps2['e00200_percentile_exact'],
[0, 25, 50, 70, 90, 99, 100],
right=False)
cps2 = cps2.drop([
'e00200_percentile_exact', 'e00200_percentile', 'e00200_2percentile',
'e00200_ventile', 'e00200_decile', 'e00200_quintile', 'e00200_quartile'
],
axis=1)
cps2 = cps2.groupby([
'income_bin', 'age_head'
]).apply(lambda x: pd.Series({
'nu18': mdf.weighted_mean(x, 'nu18', 's006'),
'n1864': mdf.weighted_mean(x, 'n1864', 's006'),
'n65': mdf.weighted_mean(x, 'n65', 's006'),
}))
cps2.reset_index(inplace=True)
cps2[cps2['age_head'].between(20, 80)]
smoothed18 = []
for group in cps2['income_bin'].unique():
temp = cps2[cps2['income_bin'] == group]
x = temp["age_head"]
y = temp["nu18"]
# plt.plot(x, y, label=group)
z18 = lowess(y, x, frac=0.4)
plt.plot(z18[:, 0], z18[:, 1], label=group)
smoothed18.extend(z18[:, 1])
def test_weighted_mean():
# TODO: Add None default to w.
# assert mdf.weighted_mean(df, 'x') == 8 / 3
assert mdf.weighted_mean(df, "x", "w") == 11 / 6
cdfs = pd.concat(cdf_list)
# Create DataFrame summarized at the UBI amount, with columns for:
# - d_stat and associated net worth
# - median and mean net worth by white/black
# - share with net worth above $50k by white/black
cdfs_max = (cdfs.sort_values("d_stat_cand",
ascending=False).groupby("ubi_mo").head(1))
ubi_summary = (simulated.groupby("ubi_mo").apply(lambda x: pd.Series({
"black_median_networth_pa":
mdf.weighted_median(x[x.race2 == "Black"], "networth_pa_new", "wgt"),
"white_median_networth_pa":
mdf.weighted_median(x[x.race2 == "White"], "networth_pa_new", "wgt"),
"black_mean_networth_pa":
mdf.weighted_mean(x[x.race2 == "Black"], "networth_pa_new", "wgt"),
"white_mean_networth_pa":
mdf.weighted_mean(x[x.race2 == "White"], "networth_pa_new", "wgt"),
"black_share_above_50k":
x[(x.race2 == "Black") &
(x.networth_pa_new >= 50000)].wgt.sum() / totals.black_hhs,
"white_share_above_50k":
x[(x.race2 == "White") &
(x.networth_pa_new >= 50000)].wgt.sum() / totals.white_hhs,
})).reset_index())
ubi_summary["white_mean_nw_as_pct_of_mean_black"] = (
ubi_summary.white_mean_networth_pa / ubi_summary.black_mean_networth_pa)
ubi_summary["white_median_nw_as_pct_of_median_black"] = (
ubi_summary.white_median_networth_pa /
ubi_summary.black_median_networth_pa)
def pov(groupby, data=person_sim):
return (
data.groupby(groupby)
.apply(lambda x: mdf.weighted_mean(x, "poverty_flag", "asecwt"))
.reset_index()
)