def mediation(df, x, m, y):
p = Process(data=df, model=4, x=x, y=y, m=[m], suppr_init=True)
y_score = p.outcome_models[y].coeff_summary()[['coeff', 'p']].round(3)
m_score = p.outcome_models[m].coeff_summary()[['coeff', 'p']].round(3)
y_score = y_score.to_dict('index')
m_score = m_score.to_dict('index')
scores = dict()
scores['x-m'] = m_score[x]
scores['m-y'] = y_score[m]
scores["x-y"] = y_score[x]
G = nx.Graph()
G.add_node(x, pos=(10, 10), name=x)
G.add_node(m, pos=(20, 13), name=y)
G.add_node(y, pos=(30, 10), name=m)
G.add_edge(
x,
y,
weight=4,
)
G.add_edge(x, m, weight=2)
G.add_edge(m, y, weight=2)
pos = nx.get_node_attributes(G, 'pos')
plt.figure(figsize=(18, 8))
plt.margins(x=0.2, y=0.2)
labels = {node: node for node in G.nodes()}
nx.draw(G,
pos,
edge_color='lightgrey',
width=1,
linewidths=1,
node_size=1300,
node_color='pink',
labels=labels,
node_shape='s',
font_size=21)
nx.draw_networkx_edge_labels(G,
pos,
edge_labels={
(x, y): scores['x-y'],
(x, m): scores['x-m'],
(m, y): scores['m-y']
},
font_color='red',
font_size=21)
plt.axis('off')
return plt
def get_ols_accuracy(model_num):
with open("Results/Results_OLS_Model{}.txt".format(model_num)) as f:
spssoutput = f.read().split(
"******************** DIRECT AND INDIRECT EFFECTS *************************\n"
)[1]
data = pd.read_csv("Data/Data_Model{}.csv".format(model_num))
varlist = open(
"Data/Varlist_Model{}.txt".format(model_num)).read().split(",")
kwargs = {i: i for i in varlist if "m" not in i}
if ("m1" in varlist) & (model_num > 3):
kwargs["m"] = ["m1", "m2"]
else:
kwargs["m"] = "m"
process = Process(data,
model_num,
precision=4,
conf=95,
modval={},
quantile=0,
logit=0,
seed=123456,
suppr_init=True,
total=1,
**kwargs)
dir = process.direct_model
indir = process.indirect_model
if dir.has_moderation:
truedir = get_direct_effect_mod(spssoutput)
else:
truedir = get_direct_effect(spssoutput)
estdir = dir.coeff_summary()
if indir.has_moderation:
trueindir = get_indirect_effect_mod(spssoutput)
else:
trueindir = get_indirect_effect(spssoutput)
estindir = indir.coeff_summary()
# Sort datasets
coldir = ["Effect", "SE", "t", "p", "LLCI", "ULCI"]
colindir = ["Effect", "Boot SE", "BootLLCI", "BootULCI"]
sortcols_dir = [i for i in truedir.columns if i not in coldir + [""]]
sortcols_indir = [i for i in trueindir.columns if i not in colindir]
if sortcols_dir:
estdir = estdir.sort_values(by=sortcols_dir)
if sortcols_indir:
estindir = estindir.sort_values(by=sortcols_indir)
direffects = np.isclose(truedir[coldir].values,
estdir[coldir].values,
rtol=1e-3,
atol=1e-3).mean()
indireffects = np.isclose(trueindir["Effect"].values,
estindir["Effect"].values,
rtol=1e-3,
atol=1e-3).mean()
indirboots = np.isclose(trueindir[colindir].values,
estindir[colindir].values,
rtol=1e-2,
atol=1e-2).mean()
return direffects, indireffects, indirboots
from pyprocessmacro import Process
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
df = pd.read_csv("./TestCases/Data_Model12.csv")
varlist = open("./Tests/Data/Varlist_Model12.txt").read().split(",")
kwargs = {i: i for i in varlist if "m" not in i}
df["ZOO"] = df["z"]
df["KA"] = df["w"]
kwargs["m"] = ["m1", "m2"]
kwargs["z"] = "ZOO"
kwargs["w"] = "KA"
p = Process(data=df, model=12, **kwargs, seed=14568743, hc3=True)
jn_regions = p.compute_jn_region("direct", "KA")
print(p.spotlight_direct_effect({"KA": [*jn_regions[0], *jn_regions[1]],
"ZOO":[0]}))
jn_regions = p.compute_jn_region("indirect", "KA", med_name="m1")
print(p.spotlight_indirect_effect("m1", {"KA": [*jn_regions[0], *jn_regions[1]],
"ZOO":[0]}))
g = p.plot_conditional_direct_effects(x="KA", modval={"KA":np.linspace(-2, 2)})
for r in jn_regions:
for v in r:
g.ax.axvline(v)
g.ax.set_xlim(-1, 1)
plt.show()