Transcript ECON 246 - Meeting #3

Urban and Regional
Economics
Week 3
Tim Bartik
“Business Location Decisions in the
U.S.: Estimates of the Effects of
Unionization, Taxes, and Other
Characteristics of States” Journal of
Busines and Econmic Statistics, Vol. 3, No.
1, Jan. 1985, pp. 14-22.
 This is a more technical article, and
hence I will present this one.

According to Bartik, what
explains locational choices of
manufacturing firms for new
branch plants?
Firms are profit maximizers, thus
expected profitability determines
locational choice.
Expected Profits depend on:
Labor market conditions.
 Other input prices

 e.g.,
energy, land, etc.
Agglomeration economies
 Fiscal conditions

 e.g.,

taxes, local subsidies, public services, etc.
First review simple logit, and then more
complex conditional logit.
Simple Logit Model





Suppose we examine a
choice to locate in
Wisconsin.
locwisc=1 if yes, and
locwisc=0 otherwise
Assume
locwisc=f(taxrate)
Only 1’s & 0’s revealed.
Need to keep
predictions in 0-1 range
Pr(locate
in WI)
1
0
1/taxrate
Logistic model uses following
functional form
ln[P/(1-P)]=B’X
 where P=prob. of Wisconsin
(ie., locwisc=1), X is set of independent
variables, B is vector of coefficients
including constant.
 This transformation keeps prediction in
0-1 range.
 Conditional logit is more complex.

Conditional Logit Model

Here we consider more than one
alternative.
 For
example, firms choosing between states
have 49 states that are alternatives to actual
choice.

Look at the application for this paper.
Empirical Approach: Conditional Logit

Again, dependent variable is not continuous.
 i.e.,
either you choose a location, or you don’t.
 We now look at multiple alternatives

Probability of locating firm i at location k.
 pr(locatei,k)=f(expected
 i,k=B’X
profitsi,k)
+ ei,k
– where X=vector of locational factors, B is a vector of
parameters, and e is a disturbance term.
– Need to compare location k with all other j locations.

Thus, pr(locatei,k)=exp(B’X)/jexp(B’X)
Some Econometric Issues

One problem is an assumption that is
made regarding the error term:
 “Independence
of Irrelevant Alternatives”
– Implies no relationship between alternatives
not chosen.

Not realistic here:
 If
profits for one southern state are higher
than a northern state, it is reasonable to
assume a neighboring southern state also is
more profitable than the northern state
Can use Nested Logit

Think of this as a hierarchical decision
process.
 You
first choose the region you are moving
to (e.g., the south) and you then choose the
specific state.

Bartik notes that a nested logit can be
estimated if one uses a set of regional
dummy variables in the conditional
logit equation.
Second Econometric Issue

We don’t have data on true alternatives (i.e.,
the sites). Rather we have data at state level.
 Uses
state-wide averages to distinguish one state
from another.
Suppose land area is proxy for number of
alternative sites.
 Important question:

 Are
all sites within the state equally probable?
Dartboard Theory

If correlation of unobserved within-state
characteristics between alternative sites in the
state is zero, then larger states have more
alternatives.
 So-called
dartboard theory.
 If you have twice the land area, you have
twice the probability of being chosen.
If correlation is one, then larger states have
no more alternatives.
 Thus: Significant land area
Dartboard!

Data

Used D&B data for 1972 and 1978.
 Looked
at all manufacturing (SIC 20-39)
 Determined plant openings, closings, acquisitions
and divestitures.
 Cross-checked for accuracy by calling firms.

Look at Table 1 for variable definitions


land area, unionization rates, work stoppages, tax rates,
road miles, existing manuf. acivity, pop density, wage rate,
education, construction costs, energy prices
Also included regional dummies
Findings: Tables 2-4

Dartboard Theory Confirmed



Large effect of unionization.


10% increase in land area increases probability of
that state being chosen by 10%
It must be the case that unobserved characteristics
within states are not correlated.
10% increase in %-unionization in state reduces
number of branch plants by 30-45%
State tax rates have expected sign.


Corp. is significant, property not quite.
Elasticity is small, but corp. tax more important
than corp. property tax.
Other Findings

Infrastructure has slight positive influence.


Existing manufacturing increases new starts.


i.e., road miles positive but elasticity approx. (0.4).
Elasticity (0.8-0.9)
High wages reduce new starts.

Elasticity (0.9)
Remaining variables insignificant.
 Added work stoppage variable not significant.



Slightly reduced unionization magnitude.
Unionization still neg. and significant and 2x work stop.
Comments?
Look at determinants of
county growth
Leichenko paper
Presented by: