In chaudhary-amit/acblm: Two step estimation of three sided heterogeniety

Simulating and estimating the mixture model

require(acblm)
require(knitr)
require(kableExtra)
options(knitr.table.format = "html") 
knitr::opts_chunk$set(dev = "png", dev.args = list(type = "cairo-png"))

Simulating a data set

In time period $t$, the distribution which depends on the type $\alpha_i$ (worker in paper), class $m_{it}$ (manager in paper) and the class $k_{it}$ (firm class in paper).

$$Pr[Y_{it} \leq y|m_{it}=m,k_{it}=k,\alpha_i=\alpha]=F_{mk\alpha}(y)$$

set.seed(3236)

# three sided model
model_test = m2.mixt.new(nk=2,nf=2,nb=2)

# model initializer for beta tests 
model <- ModelInitializer()

# assign the means of distributions system has simple complementarity 
model_test$A1[1,,]=model$A1[1,1:2,]
model_test$A1[2,,]=model$A1[1,3:4,]

# assign the variance of distribution ( small: will help in convergence on desktop)
model_test$S1[] = 0.1
model_test$S2[] = 0.1

# Simple case when distribution at T=2 is same to T=1 
model_test$A2 = model_test$A1

model_test$NNm[,,,] = 1000
model_test$pk1[,,]=0.5
model_test$pk0[,,]=0.5


pk1    = rdirichlet(2*2*2*2,rep(1,2))

dim(pk1) = c(2*2, 2*2 , 2)
model_test$pk1 = pk1


pk0    = rdirichlet(2*2,rep(1,2))
dim(pk0) = c(2,2,2)
model_test$pk0 = pk0

# simulate the data 
test_data <- Simulate.data.threeSided(model_test)

Set the model controls

ctrl <- set.solver.controls(model_test,         # Model  
                            n_startValues=1,    # number of starting values
                            stayers_sample=0.1) # if want to subsample the stayers data 

Step 1 of estimation : Clustering manager and firms

ad_employee_em <- threeSided.Clustering(test_data) # simulated data  

Step 2 of estimation : Model parameters estimation (modified EM: see paper)

my_model_test_cluster <- estimation.threeSided.model(model_test, # model 
                                                    ad_employee_em, # data with step 1 estimation results 
                                                    ctrl) # control parametrs for solver 
# Estimation results for mean                                                     
my_model_test_cluster$model$A1[1,,]

Proportion plots

# for manager class 1 
threeSided.proportion.plot(my_model_test_cluster, m=1)

# for manager class 2
threeSided.proportion.plot(my_model_test_cluster, m=2)

Estimated parameters (means)

# for manager class 1 
threeSided.means.plot(my_model_test_cluster, m=1)

# for manager class 2
threeSided.means.plot(my_model_test_cluster, m=2)

chaudhary-amit/acblm documentation built on Dec. 19, 2021, 3:02 p.m.