OOI (version 0.1.0)

OOI: Outside Option Index

Description

calculates the 'outside option index' (defined as \(-\sum P(Z|X) * log(P(Z|X) / P(Z))\)) for workers, using employer-employee data.

Usage

OOI(
  formula = NULL,
  X,
  Z = NULL,
  X.location = NULL,
  Z.location = NULL,
  wgt = rep(1, nrow(X)),
  pred = TRUE,
  method = "logit",
  sim.factor = 1,
  dist.fun = geo_dist,
  dist.order = NULL,
  seed = runif(1, 0, .Machine$integer.max)
)

Arguments

formula

a formula describing the model to be fitted in order to estimate P(Z|X) / P(Z). This formula uses a syntax similar to STATA, and so "x_" refers to all variables with the prefix "x", while "z_" refers to all variables with the prefix "z". Similarly, "d" refers to the distance polynomial (see the example below).

X

matrix or data frame with workers characteristics. Note that all column names should start with "x" (necessary for the inner function 'coef_reshape').

Z

an optional matrix or data frame with jobs characteristics. Note that all column names should start with "z" (necessary for the inner function 'coef_reshape').

X.location

an optional matrix or data frame with location for workers. Could be geographical location (i.e., geo-coordinates) or any other feature that can be used in order to measure distance between worker and job using 'dist.fun'. Currently the package supports only numeric inputs.

Z.location

same as 'X.location' but for jobs.

wgt

an optional numeric vector of weights.

pred

logical. If TRUE (default), predicts the ooi for the provided data.

method

a method for estimating P(Z|X) / P(Z). Currently not in use.

sim.factor

a variable that determines how much fake data to simulate (relative to real data).

dist.fun

a distance function to calculate the distance between X.location and Z.location. Users interested in using more than one distance metric should provide a function that returns for each row of X.location and Z.location a vector with all the necessary metrics. Also - the function should use columns by their index and not by their names. The default function is geo_dist, which is suitable for data with geo-coordinates.

dist.order

a numeric vector specifying for each distance metric an order of the distance polynomial.

seed

the seed of the random number generator.

Value

An "ooi" object. This object is a list containing the following components:

coeffs

coefficients from the estimated logit.

coeffs_sd

coefficients SE.

pseudo_r2

McFadden's pseudo-R squared for the estimated logit.

standardized_coeffs

standardized coefficients.

ooi

the Outside Option Index.

hhi

the Herfindahl-Hirschman Index, an alternative measure for outside options.

job_worker_prob

the log probability of each worker to work at his *specific* job (rahter than to work at a job with his specific z)

orig_arg

a list containing the original arguments (necessary for predict.ooi).

Examples

# NOT RUN {
#generate data
#worker and job characteristics:
n <- 100
men <- rbinom(n, 1, 0.5)
size <- 1 + rgeom(n, 0.1)
size[men == 0] <- size[men == 0] + 2
worker_resid <- data.frame(r = round(runif(n, 0, 20), 1))
job_location <- data.frame(l = round(runif(n, 20, 40), 1))
#prepare data
#define distance function:
dist_metric <- function(x, y){abs(y - x)}
X <- data.frame(men = men)
Z <- data.frame(size = size)
#add "x" / "z" to column names:
X <- add_prefix(X, "x.")
Z <- add_prefix(Z, "z.")
#estimate P(Z|X) / P(Z) and calculate the ooi:
ooi_object <- OOI(formula = ~ x_*z_ + x_*d + z_*d, X = X, Z = Z,
                  X.location = worker_resid, Z.location = job_location,
                  sim.factor = 3, dist.fun = dist_metric, dist.order = 3)
#we can extract the ooi using predict():
ooi <- predict(ooi_object)
summary(ooi)
# }