def data_clean(self, data):
""" Clean NaNs out of necessary data """
first = lu.rank_evol_vars(react=self.react)
second = ['TopRanked', 'BottomRanked', 'InsideRanked', 'OverallRank',
'Ranked', 'RankedL']
vars_keep = [first, second]
vars_keep = [item for sublist in vars_keep for item in sublist]
return data[vars_keep].dropna()
def problem(self):
""" Problem definition for separated specification """
x_vars = lu.rank_evol_vars(tilde=0, react=self.react)
rbar = np.max(self.data.loc[self.data['Ranked'] > 0, 'OverallRank'])
out = {}
## Data subsets
data_r = self.data[self.data['RankedL'] > 0]
data_rr = data_r[data_r['Ranked'] == 1]
model_t = self._r_tobit(data_rr, x_vars, rbar)
out['beta_t'] = model_t['beta']
out['sigma_t'] = model_t['sigma']
return out
def ev_tobit(self, tilde):
""" Generate expected value for interior rank """
period = self.data['year'][0] - 2013
if tilde:
beta = self.firststage['params']['rank']['beta_tilde_t']
sigma = self.firststage['params']['rank']['sigma_tilde_t']
shock = self.shock['ev_tilde'][self.active, period]
else:
beta = self.firststage['params']['rank']['beta_t']
sigma = self.firststage['params']['rank']['sigma_t']
shock = self.shock['ev'][self.active, period]
x_vars = lu.rank_evol_vars(tilde=tilde, react=self.opts['reaction'],
sim=True)
xbeta = sm.add_constant(self.data[x_vars][self.active])
xbeta = np.dot(xbeta, beta)
lambda_tobit = ((norm.pdf((1 - xbeta + shock) / sigma) -
norm.pdf((lc.RBAR - xbeta + shock) / sigma)) /
(norm.cdf((lc.RBAR - xbeta + shock) / sigma) -
(norm.cdf((1 - xbeta + shock) / sigma))))
out = xbeta + shock + sigma * lambda_tobit
return out
