-
Notifications
You must be signed in to change notification settings - Fork 1
/
03_Analysis_Appeals.py
245 lines (199 loc) · 10.3 KB
/
03_Analysis_Appeals.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
# Ryan Hübert and Ryan Copus
# Political Appointments and Outcomes in Federal District Courts
# Journal of Politics
# January 2021
## Data and supporting materials necessary to reproduce the numerical results
## in the article are available in the Journal of Politics Dataverse
## (https://dataverse.harvard.edu/dataverse/jop) and at the website of the
## corresponding author (https://ryanhubert.com/). An online appendix with
## supplementary material is available at the Journal of Politics website
## and at the website of the corresponding author.
import numpy as np
import pandas as pd
## Define working directory
## Note: you should paste the full path to your local working directory
root = ""
## Root directory must have a trailing slash
if root[-1] != "/":
root = root + "/"
################################################################################
## Define functions to calculate IPW, AIPW, and bootstrapped SEs
################################################################################
def IPW(datafr, out, trt, unadjusted = False, augmented = False, unit_wt = None):
if any(x not in datafr.columns for x in ["propensity","pred0","pred1"]):
return None
if unit_wt is None:
wt = [1] * len(datafr)
else:
wt = datafr[unit_wt]
# Unadjusted difference in means estimator
if unadjusted is True:
col1 = (wt * datafr[trt] * datafr[out]).sum()
col2 = (wt * (1 - datafr[trt]) * datafr[out]).sum()
col1b = (wt * datafr[trt]).sum()
col2b = (wt * (1 - datafr[trt])).sum()
if col1b > 0 and col2b > 0:
est = (1 / col1b) * col1 - (1 / col2b) * col2
else:
est = None
# IPW Estimator with renormalized weights, p. 39 of Glynn and Quinn (2010)
elif augmented is False:
col1 = ((wt * datafr[trt] * datafr[out]) / datafr["propensity"]).sum()
col2 = ((wt * (1 - datafr[trt]) * datafr[out]) / (1 - datafr["propensity"])).sum()
col1b = ((wt * datafr[trt]) / datafr["propensity"]).sum()
col2b = ((wt * (1 - datafr[trt])) / (1 - datafr["propensity"])).sum()
if col1b > 0 and col2b > 0:
est = (1 / col1b) * col1 - (1 / col2b) * col2
else:
est = None
# AIPW Estimator, Equation 3 of Glynn and Quinn (2010)
elif augmented is True:
col1 = (datafr[trt] * datafr[out])/datafr["propensity"]
col2 = ((1-datafr[trt]) * datafr[out])/(1-datafr["propensity"])
col3 = (datafr[trt]-datafr["propensity"])/(datafr["propensity"]*(1-datafr["propensity"]))
col4 = ((1-datafr["propensity"])*datafr["pred1"] + datafr["propensity"]*datafr["pred0"])
est = (1/(sum(wt))) * (wt * (col1 - col2 - (col3 * col4))).sum()
return est
def BootSEs(datafr, out, trt, clustervar, block = False, boots = 400, seed = 2020):
# datafr = pd.DataFrame(tmp); out = y; trt = t; clustervar = "jid_anon"
datafr["const"] = 1
bf = pd.DataFrame(columns=["boot", "effect", "outcome", "treatment", "obs",
"obs_boot", "estimator", "effect_wt"])
jwt = datafr.groupby(clustervar)["block"].count()
jwt = jwt.reset_index()
## Set the seed
rs = np.random.RandomState(np.random.MT19937(np.random.SeedSequence(seed)))
print(" ... Doing the bootstrap ... ")
N = len(datafr)
fes = set(datafr["block"])
for i in range(0, boots):
print(i)
datafr["bs"] = datafr[clustervar]
bs = np.random.choice(jwt[clustervar], size=len(jwt), replace=True)
bs = {x: len(bs[bs == x]) if x in bs else 0 for x in jwt[clustervar]}
datafr["bs"] = datafr["bs"].map(bs)
if block:
est0 = []
est1 = []
est2 = []
for b in fes:
Nb = datafr.loc[datafr["block"] == b, "bs"].sum()
est0.append((IPW(datafr.loc[datafr["block"] == b, :], out, trt,
unadjusted=True, unit_wt="bs"), Nb))
est1.append((IPW(datafr.loc[datafr["block"] == b, :], out, trt,
augmented=False, unit_wt="bs"), Nb))
est2.append((IPW(datafr.loc[datafr["block"] == b, :], out, trt,
augmented=True, unit_wt="bs"), Nb))
est0 = sum([(x[1] / datafr["bs"].sum()) * x[0] for x in est0 if not np.isnan(x[0])])
est1 = sum([(x[1] / datafr["bs"].sum()) * x[0] for x in est1 if not np.isnan(x[0])])
est2 = sum([(x[1] / datafr["bs"].sum()) * x[0] for x in est2 if not np.isnan(x[0])])
else:
est0 = IPW(datafr, out, trt, unadjusted=True, unit_wt="bs")
est1 = IPW(datafr, out, trt, augmented=False, unit_wt="bs")
est2 = IPW(datafr, out, trt, augmented=True, unit_wt="bs")
row = [i, est0, out, trt, N, datafr["bs"].sum(), "naive", ""]
bf = bf.append(pd.DataFrame([row], columns=bf.columns))
row = [i, est1, out, trt, N, datafr["bs"].sum(), "ipw_rn", ""]
bf = bf.append(pd.DataFrame([row], columns=bf.columns))
row = [i, est2, out, trt, N, datafr["bs"].sum(), "aipw", ""]
bf = bf.append(pd.DataFrame([row], columns=bf.columns))
## Reweight the bootstrap estimates as in Sherman and le Cessie (1997)
bf["effect_wt"] = (bf["obs_boot"] / bf["obs"]) ** (1 / 2) * bf["effect"]
return bf
################################################################################
## Estimate effects in Ninth Circuit
################################################################################
## Import appeals data
af = pd.read_csv(root + "USDC-APPEALS-JOP.csv", dtype = "object")
af = af.loc[:,["case_id", "pro_defendant", "maj_rep", "panel_anon", "CIRDIST", "YEAR"]]
af["block"] = af["CIRDIST"] + "_" + af["YEAR"]
## Import predictions
af1 = pd.read_csv(root + "Analysis/usca_balance_predictions.csv", dtype = "object")
## Merge appeals data and predictions
af = af.merge(af1.loc[af1["model"] == "full",["case_id","my_ensemble"]],
how = "left", on = "case_id")
af = af.rename(columns = {"my_ensemble" : "propensity"})
af = af.merge(af1.loc[af1["model"] == "0",["case_id","my_ensemble"]],
how = "left", on = "case_id")
af = af.rename(columns = {"my_ensemble" : "pred0"})
af = af.merge(af1.loc[af1["model"] == "1",["case_id","my_ensemble"]],
how = "left", on = "case_id")
af = af.rename(columns = {"my_ensemble" : "pred1"})
## Clean up variables
for i in af.columns:
if i in ["pro_defendant", "maj_rep"]:
af[i] = af[i].astype(float).astype(int)
if i in ["propensity","pred0","pred1"]:
af[i] = af[i].astype(float)
## Check no NAs
c1 = len(af.loc[af["propensity"].isnull(),"case_id"])/len(af)
c2 = len(af.loc[af["pred1"].isnull(),"case_id"])/len(af)
c3 = len(af.loc[af["pred0"].isnull(),"case_id"])/len(af)
if c1 > 0 or c2 > 0 or c3 > 0:
print("WARNING: MISSING VALUES")
af = af.rename(columns = {"case_id" : "file"}) # just to fix a bug
af["block_panel"] = af["block"] + "_" + af["panel_anon"]
################################################################################
## Estimate effects in Ninth Circuit
################################################################################
## What is the mean of the outcome variable?
print("Mean of the outcome variable: " + str(round(af["pro_defendant"].mean(),2)))
## Bootstrap panel clustered standard errors
bf = BootSEs(af, "pro_defendant", "maj_rep", "panel_anon", block = False,
boots = 1000, seed = 202007)
## Collect effects
cf = pd.DataFrame(columns=["effect", "se", "obs", "panels_treat",
"panels_control", "fe_count", "estimator"])
## Estimate unadjusted treatment effect
af1 = pd.concat([af.loc[:,["pro_defendant","maj_rep","block","panel_anon"]],
pd.get_dummies(af["block"])], axis=1)
af1 = af1.loc[:,af1.columns[3:]].mean().reset_index()
af1 = af1.rename(columns = {"index" : "block", 0 : "block_wt"})
af1["block_effect"] = pd.np.nan
for b in set(af["block"]):
mt = af.loc[(af["block"] == b) & (af["maj_rep"] == 1), "pro_defendant"]
mt = mt.mean()
mc = af.loc[(af["block"] == b) & (af["maj_rep"] == 0), "pro_defendant"]
mc = mc.mean()
af1.loc[af1["block"] == b, "block_effect"] = mt - mc
est = round((af1["block_wt"] * af1["block_effect"]).sum(),4)
se = round(np.sqrt(bf.loc[bf["estimator"] == "naive","effect_wt"].var()),4)
toadd = [est, se, len(af),
len(af.loc[af["maj_rep"]==1,:].groupby("panel_anon")["file"].count()),
len(af.loc[af["maj_rep"]==0,:].groupby("panel_anon")["file"].count()),
len(set(af["block"])),
"unadjusted"]
toadd = {cf.columns[x] : [toadd[x]] for x in range(0,len(cf.columns))}
cf = cf.append(pd.DataFrame(toadd))
print("Unadjusted Treatment Effect: " + str(est))
print("Unadjusted Treatment Effect (Std. Error): " + str(se))
## Estimate IPW
est = round(IPW(af, "pro_defendant", "maj_rep", augmented=False),4)
se = round(np.sqrt(bf.loc[bf["estimator"] == "ipw_rn","effect_wt"].var()),4)
toadd = [est, se, len(af),
len(af.loc[af["maj_rep"]==1,:].groupby("panel_anon")["file"].count()),
len(af.loc[af["maj_rep"]==0,:].groupby("panel_anon")["file"].count()),
len(set(af["block"])),
"ipw"]
toadd = {cf.columns[x] : [toadd[x]] for x in range(0,len(cf.columns))}
cf = cf.append(pd.DataFrame(toadd))
print("IPW Treatment Effect: " + str(est))
print("IPW Treatment Effect (Std. Error): " + str(se))
## Estimate AIPW
est = round(IPW(af, "pro_defendant", "maj_rep", augmented=True),4)
se = round(np.sqrt(bf.loc[bf["estimator"] == "aipw","effect_wt"].var()),4)
toadd = [est, se, len(af),
len(af.loc[af["maj_rep"]==1,:].groupby("panel_anon")["file"].count()),
len(af.loc[af["maj_rep"]==0,:].groupby("panel_anon")["file"].count()),
len(set(af["block"])),
"aipw"]
toadd = {cf.columns[x] : [toadd[x]] for x in range(0,len(cf.columns))}
cf = cf.append(pd.DataFrame(toadd))
print("AIPW Treatment Effect: " + str(est))
print("AIPW Treatment Effect (Std. Error): " + str(se))
## Print additional data for the table.
print("Observations: " + str(len(af)))
print("Majority Republican Panels: " + str(len(af.loc[af["maj_rep"]==1,:].groupby("panel_anon")["file"].count())))
print("Majority Democratic Panels: " + str(len(af.loc[af["maj_rep"]==0,:].groupby("panel_anon")["file"].count())))
print("Division-Years: " + str(len(set(af["block"]))))
cf.to_csv(path_or_buf= root + "Analysis/effects_appeals.csv", mode="w", index=False, header=True)