Economic Modelling 102 (2021) 105582

Contents lists available at ScienceDirect

Economic Modelling
journal homepage: www.journals.elsevier.com/economic-modelling

Is there a wage curve with regional real wages? An analysis for the US
and Poland
Bartlomiej Rokicki a, *, Uwe Blien b, Geoffrey J.D. Hewings c, Phan thi Hong Van b
a
University of Warsaw, Faculty of Economic Sciences, Poland
b
Institut für Arbeitsmarkt- und Berufsforschung (IAB), Germany
c
Regional Economics and Applications Laboratory, The University of Illinois at Urbana-Champaign, USA

A R T I C L E I N F O A B S T R A C T

JEL classification: This paper reassesses the wage curve applying regional price deflators, whereas in the literature only uniform
C26 national price levels are generally used in wage curve regressions. Existing estimates for the unemployment
J30 elasticities vary across countries but lie mostly in the neighborhood of
0.1. Using the data for the 49 US states
J60
over the 2008–2015 period and 16 Polish NUTS2 regions over the 2000–2015 period, we test the impact of
Keywords: regional prices on the wage curve employing a variety of approaches. We confirm the existence of wage curves for
Wage curve
both countries. Following the theory of monopsonistic competition, this inverse relationship stems from the costs
Spatial spillovers
Regional price deflators
faced by both firms and workers. However, the elasticity of local unemployment and spatial spillovers decreases
Regional labor markets significantly if regional price deflators are applied. This is due to the fact that there exists a negative relationship
between regional prices and unemployment.

1. Introduction Following the seminal contributions of Blanchflower and Oswald (1994),

In the labor market literature, regional price differences are normally
disregarded, although theoretical arguments are derived on the basis of
real wages. When analyzing labor demand, for example, the price level is
assumed to be the same in the whole economy (e.g., Borjas, 2013). It can
be demonstrated, however, that larger countries in particular exhibit
significant regional price differentials. In view of this finding, it is
possible that many of the empirical analyses concerning the national
labor market may be misspecified because they do not account for an
important source of intranational heterogeneity. Where attention has
been directed to the problem, justification for the use of wages without a
regional price index has been addressed through the inclusion of regional
fixed effects. However, this modification can only take account of
regional price differences if these effects are sufficiently constant over
time.
Wage curve studies may be regarded as a perfect example of possible
misspecification in the case of empirical labor market analyses. Here, the
wage curve is a nonlinear, negative relationship between individual
wages and regional unemployment rates, where the unemployment rate
can be interpreted as a measure of job competition (e.g., Longhi, 2012).
a wage curve can be explained by efficiency wage and wage negotiation
approaches. In the case of low unemployment rates and with efficiency
wages, firms have to pay higher wages to motivate workers to put rela-
tively high effort into their work. In accordance with negotiation
approach, when unemployment rates are high, workers do not dare fight
for higher wages because they fear being terminated from employment.
In addition to these approaches, one can refer to the theory of
monopsonistic competition (see Manning, 2003). In this view, the in-
verse relationship between wages and unemployment stems from the
costs faced by both firms and workers. On the one hand, firms entering a
local market with an elevated unemployment rate will not offer high real
wages, as they cope with costs related to the recruitment and training
process. On the other hand, workers are exposed to costs associated with
job searching, commuting or migration. As a consequence, they accept
lower real wages in areas with higher unemployment. Further, the local
real wage is also influenced by labor market conditions in surrounding
areas (see, Longhi et al., 2006).1
In the classical formulation by Blanchflower and Oswald (1994), who
viewed the wage curve as an “empirical law of economics,” even the size
of the coefficient was included. After analyzing the economies of 12

* Corresponding author. University of Warsaw, Faculty of Economic Sciences, Dluga 44/50, 00-241, Warsaw, Poland.
E-mail address: brokicki@wne.uw.edu.pl (B. Rokicki).
1
Note, that at the aggregate level, the relationship between real wages and unemployment rate may depend on the type of the shock applied to the economy or its
trade openness (e.g., Rhee and Song, 2020).

https://doi.org/10.1016/j.econmod.2021.105582
Received 9 June 2020; Received in revised form 7 June 2021; Accepted 16 June 2021
Available online 22 June 2021
0264-9993/© 2021 Elsevier B.V. All rights reserved.
B. Rokicki et al.
countries, they concluded that the coefficient is regularly close to
0.1.
In the double logarithmic specification of the wage curve, this implies
that a doubling of the unemployment rate would reduce wages by 10%.
Later empirical studies found a greater variety of coefficients because the
size of the effect is related to the institutional structure of the respective
economies, e.g., whether wage bargaining is centralized or not. This is
one of the reasons why the wage curve is still used in new analyses in a
number of countries (see Baltagi et al., 2017; Johansen et al., 2019; Blien
and Phan, 2019).
Wage curve estimations in the style of Blanchflower and Oswald
(1994, 2005) included fixed effects to control for regional heterogeneity,
for instance for differing regional prices. However, due to a lack of data
on regional prices, it is not clear whether price differences between re-
gions are stable or dynamic. Is it possible that the wage curve may simply
be an illusion since it vanishes when a regional price index is used? A
systematic assessment of the many studies published on the wage curve
show that very few of them include regional prices in the analysis. The
exceptions include papers by Sanroma and Ramos (2005) and Ramos
et al. (2010). Their results are important since they found a wage curve
even when including regional price indices. However, the aim of these
papers was not to systematically compare wage curves with real and
nominal regional wages. Furthermore, the price indices they use have
some limitations. The first paper includes only a proxy for regional prices
as an additional explanatory variable. The second paper covers selected
metropolitan areas only.
We investigate whether there is a significant difference between the
results of the wage curve analysis based on nationally adjusted data and
those based on regionally adjusted data. Note that throughout the paper,
we understand “nationally adjusted wages” to be the wages expressed in
constant prices in the time dimension (applying a national CPI deflator)
that do not take regional price differentials into account. “Regionally
adjusted wages”, in contrast, are expressed in constant prices that addi-
tionally take account of regional price differentials. As a consequence, we
divide wages expressed in constant prices by a regional price deflator that
takes the value greater than 1 for regions with price level higher than
average and the value lower than 1 for regions with price level lower than
average. This is important, since prices can vary over space and through
time.
By omitting regional price indices, it is possible that many papers on
the wage effects of regional unemployment that have since been pub-
lished could be incorrectly specified. Therefore, we are interested in
testing the relationship between regional unemployment and wages with
and without regional price indices. We expect that there exists a negative
correlation between regional prices and unemployment, drawing on
Vermeulen and van Ommere (2009). As a result, we expect to find that
standard wage curve analyses, based on nationally adjusted wages,
overestimate the negative effect of the regional unemployment rate on
wages. To this end, we replicate the standard approach to the estimation
of the wage curve. We also anticipate that a similar effect will occur when
wages and spatial spillovers of unemployment rates are correlated.
The remainder of the paper is organized as follows. The next section
presents the challenge associated with switching from nominal to real
prices. Section 3 reviews prior literature on regional price differentials,
while in section 4 we discuss the research methodology and the data used
in our analysis. In section 5, we present empirical results for both the US
and the Polish wage curves based on nationally adjusted as well as
regionally adjusted wages. Finally, section 6 provides some concluding
remarks.
2. The challenge
In this paper, we test whether estimations of the wage curve are
influenced by the inclusion of regional prices. Apart from the possibility
of the wage curve vanishing when regional price differences are
included, the size of the coefficient in the wage curve equation might also
be affected. This is due to the fact that while the wage curve approach
2
Economic Modelling 102 (2021) 105582
describes the relationship between the regional unemployment rates and
real wages, existing studies assume prices to be equal across regions. As a
result, there are only a few studies that effectively examine the rela-
tionship between unemployment rates and real wages, where the wages
are adjusted by fully developed regional price deflators. Blanchflower
and Oswald (1995) are aware of the potential criticism associated with
omitting regional price indices. It is necessary to bear in mind that the
negative relationship between wages and unemployment rates is stronger
the lower the wages are in regions with high unemployment and the
higher they are in regions with low unemployment. Therefore, any
negative correlation between regional unemployment rates and prices
may lead to a bias in the wage curve estimation. Indeed, the unem-
ployment rate is considered to be one of the determinants of regional
prices (see Blien et al., 2009). Moreover, Vermeulen and van Ommeren
(2009) show that there exists a negative relationship between unem-
ployment and local prices. In particular, they observe this phenomenon
in the case of the housing market. Thus, lower unemployment rates lead
to higher prices, while higher unemployment is associated with lower
price indices. Consequently, in regions with higher unemployment rates,
real wages are likely to be higher compared to nominal wages. Exactly
the opposite applies for real wages in regions with lower unemployment
rates: on average they should be lower than nominal wages. As argued by
Vermeulen and van Ommeren (2009), this compensating mechanism
discourages many workers from migrating and may explain the persis-
tence of high unemployment rates in certain areas.
Blanchflower and Oswald (1994) claim that the omission of regional
price indices in wage curve analysis is not a serious problem since, in the
case of Great Britain, the wage curve remains intact once controls for
regional prices are included. However, the data on regional prices they
use in a robustness check come in the form of a cost-of-living index
provided by a commercial company. Moreover, they cover only a very
short period of time and a limited number of regions. The authors
apparently do not regard regional price differences as an important issue,
as they neither explain the methodology used to calculate the price index
nor do they reveal the size of the regional price differentials. It is
therefore impossible to assess the data reliability and its possible impact
on the wage curve estimations. Hence, if we assume that the regional
price differentials within certain countries are sizable and dynamic, it is
more than likely that the existing estimates of the wage curve are
significantly biased since they overestimate both the negative relation-
ship between low wages and high unemployment and that between high
wages and low unemployment. The question is, how large is this bias and
is it statistically significant?
This paper provides new evidence concerning the possible over-
estimation of the wage curve. We apply regional price parities (RPP) for
the US states and Polish NUTS2-level regions.2 Poland and the US are two
of the rare cases of countries for which regional price deflators are
available. Furthermore, to the best of our knowledge, these are the only
cases with full regional coverage and longer time series of price de-
flators.3 They are sufficiently different from each other to permit gen-
eralizations of the results.
In the empirical part of the paper, we analyze local and spatial ap-
proaches. The estimates we refer to as “local” are based on equations that
include only the unemployment rate of the region in which a respective
worker is located, whereas the spatial approach also includes spillover
effects of the unemployment rates in the surrounding regions. In addi-
tion, we assess whether the results hold not only for the whole sample but
also for subsamples (e.g., males versus females). As a robustness check,
we apply the two-step approach developed by Bell et al. (2002).
2
NUTS (Nomenclature of Territorial Units for Statistics) - is a geographical
nomenclature subdividing the economic territory of the European Union into
different regional levels.
3
See Weinand and von Auer (2020) for a detailed description of existing
studies on regional price differentials.
B. Rokicki et al.
3. Regional price differentials – literature review
For many years, the existence of regional price differentials attracted
little interest in the economics literature. Deller et al. (1996), for
instance, claim that “most of the discussion of regional income conver-
gence in recent decades has operated under the assumption, either
explicit or implicit, that spatial and temporal differences in regional
prices (i.e., cost of living and inflation) are negligible.” The main reason
is probably the problem associated with data availability. For many
years, it was not possible to estimate reliable regional price indices.
In the case of the US, the first attempts to analyze regional price
differences were made in the 1970s, with authors such as Haworth and
Rasmussen (1973) relying on regression analysis to investigate the
sources of geographical differences in cost of living. These early studies
were followed by papers by Cebula (1980), Hogan (1984) and McMahon
(1991). The variables most commonly used as determinants of regional
cost-of-living differentials include per capita income, total population,
population density, housing prices, geographic area, educational attain-
ment of inhabitants and dummy variables for coastal locations or for the
existence of right-to-work legislation (e.g., Cebula, 1980). More recently,
the Bureau of Economic Analysis released, as official statistics, regional
price parities for US states and Metropolitan areas (see Aten and Fig-
ueroa, 2014). These regional price indices are estimated using hedonic
regression models (e.g., Aten and D'Souza, 2008) and are available for
the 2008–2013 period.
Different econometric approaches concerning regional price differ-
entials have also been applied elsewhere. For instance, Roos (2006),
Kosfeld et al. (2008) and Dreger and Kosfeld (2010) used regression
analysis, while Blien et al. (2009) applied multiple imputation techniques
to estimate regional price indices for Germany at different levels of ter-
ritorial aggregation. Variables correlated with prices include per capita
GDP, average wages, total population and population density, region
size, price of building land, unemployment rate, as well as dummy var-
iables for tourism and eastern Germany.
The main problem with regression-based regional price indices is
their reliability. For example, Hogan (1984) criticizes the conclusions
drawn by Cebula (1980) due to the fact that his regional cost-of-living
indices were based on aggregate budget data. He claims that the
regression model applied by Cebula (1980) was only able to “explain”
slightly more than half of the total variance in living costs for one out of
nine different consumption categories. Blien et al. (2009) argue in their
paper that the price index developed by Roos (2006) ignores the uncer-
tainty of the estimation process because it is based on the results of OLS
regressions. Several attempts have been made recently to calculate
regional price deflators using raw regional price data. As a result,
non-regression-based regional price indices have been estimated for


China (e.g., Brandt and Holz, 2006), the Czech Republic (e.g., Cadil et al.,
2014), Poland (e.g., Rokicki and Hewings, 2019) or the US (Beraja et al.,
2019). In the latter case, the regional price index is based primarily on
food products; its representativeness is therefore limited.
Are regional price differentials really sizable and can they signifi-
cantly influence regional income measures such as per capita GDP or
wages? McMahon and Melton (1978) develop a cost-of-living index for
the US states (updated by McMahon, 1991) and show that differences in
living costs are substantial – the index for 1973 ranges between 127.8 in
Connecticut and 90.6 in West Virginia (with the national average ¼ 100).
Official statistics provided by the Bureau of Economic Analysis confirm
the above findings. In 2013, regional price parities at the state level range
between 117.7 in the District of Columbia and 86.8 in Mississippi. The
differences are even larger at the metropolitan area level – the lowest
price index is found in Beckley, WV (78.0), the highest in Urban Hono-
lulu, HI (122.5). Several other studies reveal the existence of significant
regional price differentials in Europe or China. Moreover, the differences
in regional price indices seem to be persistent over time. Existing studies
show either no convergence of regional prices (e.g., Dreger and Kosfeld,
2010, in the case of German districts; Nagayasu, 2011, in the case of
3
Economic Modelling 102 (2021) 105582
Japanese regions) or very slow rates of convergence (e.g., Cecchetti,
2002, in the case of the US cities).
Several studies confirm that the results of analyses based on regional
income measures may be significantly influenced by using regional price
indices. Deller et al. (1996) apply indices developed by McMahon and
Melton (1978) and McMahon (1991) in their analysis of regional income
convergence across US states between 1969 and 1991. They show that
regional price differentials can partly “explain” the absolute level of
regional income inequality. Furthermore, they find the 1980s divergence
pattern is not supported once the nominal data are adjusted by means of
regional price deflators. Kosfeld et al. (2008) claim that regional income
disparities at the NUTS3 level in Germany are reduced significantly when
regional price differentials are taken into account. They also show that
regional price deflators influence the convergence process - real
(adjusted) per capita income converges more rapidly with the unique
steady state than nominal (unadjusted) per capita income. Rokicki
(2015) confirms significant correlations of regional price indices and
wages. He finds that application of regional price deflators decreases the
overall level of wage disparities across Polish NUTS2 regions although it
does not change the overall pattern of their evolution to any significant
extent.
Given these findings, it seems reasonable to believe that the appli-
cation of regional price indices may indeed significantly influence the
results of any wage curve analysis. Note also that the results of
regression-based studies show that income variables (e.g., per capita GDP
or wages) are positively correlated with the regional price level. This
implies that better developed areas have higher price levels on average.
As a consequence, there must be a relationship between regional prices
and unemployment (e.g., Zeilstra and Elhorst, 2014). The prior suggests
that we may face an omitted-variable bias problem given the fact that our
main explanatory variable is likely to be correlated with regional prices.
Hence, as noted earlier, we can expect the existing estimates of the wage
curve to overestimate both the negative relationship between low wages
and high unemployment and that between high wages and low
unemployment.
This paper adds to the existing literature by discussing the impact of
regional price differentials on results obtained in wage curve analysis. In
particular, it compares the results of wage curve analysis for the US and
Poland, based on nationally adjusted wages (deflated only in time by the
average/national price deflators) and regionally adjusted wages that take
into account regional price differentials. We find that, in both countries,
the local unemployment elasticity and spatial spillovers decrease signif-
icantly once regional price deflators are applied. In some cases, the
application of regional price deflators causes the wage curve to become
statistically insignificant.
4. Research methodology and data
4.1. Methodology
Since the initial study by Blanchflower and Oswald (1990), the
negative relationship between wages and local unemployment rates has
been the subject of many papers. Blanchflower and Oswald (2005) claim
that wage curve studies cover over 40 countries and that in most of the
cases the estimates of unemployment elasticities exhibit values around

0.1. However, this result concerning the magnitude of the effect is
questioned by Nijkamp and Poot (2005). After conducting a
meta-analysis, they claim that, due to publication bias, the average wage
curve elasticity is no more than
0.07. Thus, the wage curve results and
estimation methodology are not exempt from controversy. Card (1995),
for example, suggests the need to apply subsample-specific unemploy-
ment rates, since results based on average unemployment rates can be
biased (downwards) for certain subsamples. In addition, Longhi et al.
(2006) claim that in the presence of monopsonistic competition, em-
ployers are likely to be aware of the employment opportunities in sur-
rounding areas. The local unemployment elasticity is therefore likely to
B. Rokicki et al.
be overestimated if the existence of spatial spillovers is not taken into
account. Therefore, the wage curve estimates should control for both
region heterogeneity and spatial spillovers in order to avoid estimation
bias and misleading inference. Note that on average the magnitude of
spatial spillover decreases with an increasing scale of spatial units (e.g.,
Chung and Hewings, 2019). In our case, this should result in a spillover
elasticity that is lower for the US states and higher for the considerably
smaller Polish NUTS2 regions.
While there have been several methodological improvements, none of
the recent studies takes into account possible overestimation bias
resulting from the omission of regional price deflators. The results con-
firming a significant bias could be particularly helpful in explaining high
estimates of the spatial spillovers found in the studies by Fingleton and
Palombi (2013) or Baltagi and Rokicki (2014). The first study finds that
the unemployment rate within commuting distance exerts the strongest
influence on wages, while the local unemployment rate accounts only for
a small proportion of the total impact. The second paper shows that the
spatial spillovers in unemployment rates more than double the elasticity
of the local unemployment rate.
According to Blanchflower and Oswald (1990), the wage curve is a
standard wage equation commonly used to estimate returns to education
or the male–female wage gap, but with the local unemployment variable
added to the set of regressors. A set of regional fixed effects is included to
control for regional heterogeneity. In our case, in order to avoid esti-
mation bias and misleading inference (see Longhi et al., 2006), we also
include spatial spillovers of the regional unemployment rate. The equa-
tion estimated takes the following form4:
!

X 0
log Wirt ¼ α þ β log Urt þ χ Grs log Ust þ Xirt γ þ μr þ λt þ νirt
s NUTS2-level regions (see Rokicki and Hewings, 2019) and microdata
where Wirt is the real hourly wage rate of individual i observed in region r
at time t. Urt is the unemployment rate in region r at time t. μr is a region
effect, λt is a time effect and νirt is the remaining error term. Xirt represents
control variables that include the characteristics of individual i such as:
gender, age, age squared, tenure, tenure squared, education, marital
status, occupation, industry, sector, size of the employing firm, job tenure
and city size. The matrix G is a row-normalized spatial weights matrix
(meaning that each row sums to one) that symbolizes the connections
between regions r and s with the typical element Grs where r, s ¼ 1, 2,..,R.
G is of dimension RR with diagonal elements equal to zero. The
P
spatially weighted unemployment regressor ( s Grs log Ust ) allows the
unemployment rates of the neighboring regions to influence local wages.
Longhi (2012) shows that the unemployment rate could be substituted by
a different measure of job competition. However, we retain the original
measure introduced by Blanchflower and Oswald.
As a robustness check, we also test the wage curve with regional
prices in a framework using the two-step approach suggested by Card
(1995) and applied by Bell et al. (2002), Baltagi et al. (2012) and others.
The advantage of the two-step approach is that the two crucial variables,
i.e., wages and unemployment, are measured at the same level of ag-
gregation. Furthermore, the standard errors represent observations of
regions and not observations of individuals. An instrumental variable
approach for both models (specified for individuals and for regions) is
intended to exclude reverse causality. In the 2SLS estimations, the
temporally lagged unemployment rates are used as instruments.
4
We do not include spatial and temporal lags because the wage curve liter-
ature normally avoids these extensions. The standard wage curve (following
Blanchflower and Oswald, 1990) is formulated without these lags, whereas they
are included in several papers addressing special research questions. We intend
to be comparable to the standard formulations in the literature. The second
reason for avoiding a model with lags is that the spatial lag could not be applied
in models with individual data and we intend to use the same basic model with
pure individual data and with our two-step approach.
Economic Modelling 102 (2021) 105582
In the two-step approach, first a region/time-specific fixed effect αrt is
calculated in a sequence of cross-sections:
0
log Wirt ¼ αt þ αrt þ Xirt γ þ νirt (2)
The fixed effect αrt can be regarded as a regional wage that is stripped
of all the individual characteristics of the local labor force. In a second
step, the real wage curve is calculated by using the composition corrected
wage αrt as the response variable:
!
X
0
αrt ¼ β0 þ β log Urt þ χ Grs log Ust þ μ0r þ λ0t þ ν0irt (3)
s
4.2. The data
Recently, the Bureau of Economic Analysis released regional price
parities for US states as official statistics, covering the period between
2008 and 2015 (e.g., Aten and Figueroa, 2014). We thus focus on the US
and use microdata from the US census to analyze the impact of regional
price differentials on the results of wage curve analysis. In order to check
for the robustness of our findings, we also assess the impact of regional
price deflators on wage curve estimation in Poland. To the best of our
knowledge, Poland is the only European country where such data have
been available for a number of years. A regional price index has been
recently developed for Germany (Weinand and von Auer, 2020). How-
ever, as the data refer only to a cross-section, they cannot be used in a
wage curve analysis. There is a similar problem with the data for the
Czech Republic (Cadil et al., 2014) and the United Kingdom (ONS, 2018).
In this paper, we apply regional price deflators developed for Polish
(1)
from the Labor Force Survey for the 2000–2015 period.
The regional price parities (RPP) for the US states come from the
Bureau of Economic Analysis, US Department of Commerce. The indi-
vidual data on wages are from the US census (see Ruggles et al., 2015) –
our sample includes over 7 million observations at the state level.5 Since
we wish to analyze both the standard and the spatial wage curve, we
exclude Alaska and Hawaii. The aggregated data used in the two-step
approach comprise 392 observations (49 regions times 8 years). The
data on unemployment come from the Bureau of Labor Statistics, US
Department of Labor while the data on regional population are from the
US Census Bureau. The housing price index is obtained from the Federal
Housing Finance Agency and the data on land area are from the SAGE
Stats. All data cover the 2008–2015 period. The data used to compute
spatial weights matrices come from the shapefile data for US regions.
The regional price deflators for Poland are estimated in accordance
with the conventional Eurostat/OECD methodology (see European
Communities/OECD, 2006, and Rokicki and Hewings, 2019 for more
details). Almost all of the remaining data are obtained from various
publications of the Polish Central Statistical Office. In particular, the data
on regional unemployment rates are from Local Data Bank, while indi-
vidual wage data are from the Polish Labor Force Survey. In the case of
the latter, our sample includes 142,010 observations at the NUTS2 level.
The aggregated data used in the two-step approach comprise 256 ob-
servations (16 regions times 16 years). All data cover the 2000–2015
period. The data used to compute the spatial weights matrix come from
the ESRI shapefile data for Polish NUTS2 regions.
5. Empirical results
All the results reported in this paper are based on an instrumental
variables approach that deals with possible endogeneity of the unem-
ployment rate. In our baseline estimations, we use the lagged value of the
unemployment rate as an instrument and present FE-2SLS estimates. We
5
See Appendix for summary statistics on our US and Polish microdata.
4
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Table 1
Verification of omitted variable bias – US states and Polish NUTS2 regions (response variable - nominal hourly wages).
VARIABLES US states Polish NUTS2 regions

(1) (2) (3) (1) (2) (3)

Model Model Model Model Model Model

Unemployment
0.049***
0.034*** 0.006***
0.067***
0.059***
0.083***

(0.005) (0.005) (0.001) (0.012) (0.012) (0.006)
RPP deflator 0.581*** 0.644*** 0.469*** 1.063***
(0.033) (0.002) (0.096) (0.031)
Constant 1.741*** 1.790*** 1.723*** 2.515*** 2.488*** 2.541***
(0.010) (0.011) (0.004) (0.068) (0.068) (0.061)

Region dummies Yes Yes No Yes Yes No

Year dummies Yes Yes Yes Yes Yes Yes

Observations 7,245,942 7,245,942 7,245,942 142,010 142,010 142,010

R-squared 0.38 0.38 0.38 0.47 0.47 0.47
Adj. R-squared 0.38 0.38 0.38 0.47 0.47 0.47

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate by region in the previous year was used as an instrument for the logarithm of the unemployment rate by
region at time t.

begin our analysis by demonstrating the existence of omitted variable
bias. Table 1 compares the results of local wage curve estimations based
on equation (1) without spatial spillovers for the US states and Polish
NUTS2 regions. As expected, in both cases, we find the RPP variable to be
positively and statistically significantly correlated with nominal wages.
Moreover, the unemployment elasticity decreases once RPP are included
(model 1 versus model 2).
Note, that both model 1 and model 2 include regional fixed effects
that are supposed to absorb regional price differences if these are con-
stant over time. In fact, when we exclude regional fixed effects (model 3)
the RPP coefficient increases. In the case of Polish NUTS2 regions it is
very close to 1. The hypothesis concerning the existence of fluctuations in
regional price indices can be further confirmed once we analyze their
evolution across time. Fig. 1 shows that, both in the case of the US states
and Polish NUTS2 regions, there exists a non-negligible wavering of the
RPP.
Next, we turn to the wage curve estimation based on nationally
adjusted and regionally adjusted wages. We start with the local wage
curve - the wage curve without spatial spillovers. Table 2 shows the
estimation results of equation (1) for the US states based on the fixed
effects approach, controlling for individual worker characteristics. We
test two specifications: in the first one, the wage variable expressed in
constant prices does not take regional price differentials into account
(“nationally adjusted”), while in the second one, wages are additionally
deflated using RPP (“regionally adjusted”). We therefore compare the
same model with two different dependent variables.
The first two columns of Table 2 report the results for the whole
sample. As expected, the unemployment elasticity seems to be signifi-
cantly lower for the specification based on regionally adjusted wages
(
0.023) as compared to the usual specification that does not take dif-
ferences in regional price indices into account (
0.049). The confidence
bands of the coefficients do not overlap. We thus conclude that there is a
significant difference between the two estimates, implying that the un-
employment elasticity is in fact considerably lower when nationally
adjusted wages are replaced by regionally adjusted wages.
It has been argued (e.g., Baltagi and Rokicki, 2014) that males and
females generally compete for different jobs, so their situation in the
labor market also differs. The same applies, for example, to workers with
different levels of educational attainment. As a result, the question arises
as to whether the magnitude of overestimation is similar across different
groups on the labor market. If it is not, then local price levels must have
differing impacts on nominal wages of different groups of workers.
Therefore, in columns 3–4 and 5–6, we report the results for the male and
female subsamples. Here, although the unemployment elasticities for
men and women differ significantly (
0.062 vs.
0.039 in the case of the
nationally adjusted wages) the impact of regional price deflators appears
to be quite similar. The application of the regional price deflators, based
on the point estimates, results in the unemployment elasticity decreasing
by
0.026 in the case of males and
0.025 in the case of females. This
implies a reduction in the unemployment elasticity by around 40% for
men and more than 60% for women. Yet, the sign and statistical signif-
icance of the relationship between wages and local unemployment rate
are not changed by applying regional price deflators.
Table 3 shows the results of a similar analysis performed for Polish
NUTS2-level regions. Here, we find similar patterns to those of the US
states. However, the negative relationship between wages and unem-
ployment is stronger: the unemployment elasticity for whole sample
based on nominal wages is
0.067. Furthermore, the impact of the
regional price indices is smaller: the unemployment elasticity for the
specification based on regionally adjusted wages reaches
0.051, a
reduction of less than 25%. Similar results are found for the male and
female subsamples. In the case of males, the unemployment elasticity
decreases from
0.077 to
0.063, which implies a reduction of almost
20% while, for females, the decline is larger, reaching almost 30% (from

0.058 to
0.041).
A problem with wage curve analyses is the potential endogeneity of
the unemployment rate. To deal with the above issue, we instrument this
key variable following the approach found in the wage curve literature.
Tables 2 and 3 show the second-stage results of 2SLS regressions in which
unemployment rates are instrumented by their first lags. First-stage es-
timates for both the US and the Polish wage curve show that the lagged
values are correlated with the explanatory variable and pass the weak
identification test. Moreover, the usual diagnostics6 indicate that while
the unemployment rate should be treated as endogenous in the US case, it
can be considered exogenous in the Polish case.7 To make sure that we
can exclude endogeneity, we use instruments in the analyses of both
countries. Given that the number of excluded instruments equals the
number of endogenous regressors, we cannot test the hypothesis
6
We apply an endogeneity test defined as the difference of two Sargan-
Hansen statistics: one for the equation with the smaller set of instruments,
where the suspect regressor is treated as endogenous, and one for the equation
with the larger set of instruments, where the suspect regressors are treated as
exogenous. Under the null hypothesis, the specified endogenous regressors can
actually be treated as exogenous.
7
In order to save space, we report the endogeneity test only. However, first-
stage estimates are available from the authors on request.

5
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Fig. 1. Evolution of regional price indices in the US and Poland (national average ¼ 1).

concerning the exogeneity of our instrument. However, as shown by equation, and if they are sufficiently correlated with the explanatory
Reed (2015), the use of the lagged values of the endogenous variable as variable that is determined simultaneously.
instruments in 2SLS/GMM/LIML estimations can be an effective strategy We have also conducted further sensitivity tests. They consisted of
if the lagged values do not themselves belong in the respective estimating additional 2SLS regressions based on the Bartik (1991) measure8 for the
US and additional OLS regressions for Poland. In the case of the former,
we found that applying the Bartik measure as an instrument for the un-
8
employment rate actually increases the difference between the unem-
The Bartik (1991) measure has been widely used in the labor market liter-
ployment elasticity of nationally adjusted and regionally adjusted wages
ature as an instrument for unemployment rate (e.g., Blanchard and Katz, 1992;
(see Table A1 in the Appendix). At the same time, the results of the OLS
Wozniak, 2010). It is a measure of local labor demand that is not related to
changes in the local labor supply. In our case it takes the following form:
estimations for Poland are very similar to the 2SLS results and do not
change our findings concerning the impact of RPP on wage curve esti-
X
j   mates (see Table A2 in the Appendix).
^ ^
Bartikst ¼ θsjt
1 lnE jt
lnE jt
1
j¼1
As an additional robustness check, we apply the two-step approach
developed by Bell et al. (2002), which is described by equations (2) and

jt
where θsjt
1 is the share of employment in industry j of state s in year t-1 and lnE (3). Table 4 reveals the second-step estimates for the US states. On
is the natural log of national employment (excluding the employment of state s) average, the results are very similar to those based on the commonly used
in industry j in year t. The data used to calculate the Bartik measure come from microeconometric approach. Unemployment elasticities are only slightly
the Bureau of Economic Analysis and cover the period 2007–2015.

6
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Table 2
The US unemployment elasticity of hourly wages – the impact of regional PPP deflators on the local wage curve.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.049***
0.023***
0.062***
0.036***
0.039***
0.014**

(0.005) (0.005) (0.007) (0.007) (0.007) (0.007)
Constant 1.741*** 1.826*** 1.533*** 1.618*** 1.631*** 1.717***
(0.010) (0.010) (0.013) (0.013) (0.014) (0.014)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Endogeneity test P-val 0.000 0.000 0.000 0.000 0.000 0.000

Observations 7,245,942 7,245,942 3,667,084 3,667,084 3,578,858 3,578,858

R-squared 0.38 0.38 0.39 0.39 0.38 0.37
Adj. R-squared 0.38 0.38 0.39 0.39 0.38 0.37

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate by state in the previous year was used as an instrument for the logarithm of the unemployment rate by
state at time t.
c) Apart from regional and time fixed effects we use a set of control variables such as age, gender, race, marital status, educational attainment, occupation and industry.
d) The endogeneity test is defined as the difference of two Sargan-Hansen statistics: one for the equation with the smaller set of instruments, where the suspect regressor
is treated as endogenous, and one for the equation with the larger set of instruments, where the suspect regressors are treated as exogenous. Under the null hypothesis,
the specified endogenous regressors can actually be treated as exogenous.

Table 3
The Polish unemployment elasticity of hourly wages – the impact of regional PPP deflators on the local wage curve.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.067***
0.051***
0.077***
0.063**
0.058***
0.041**

(0.012) (0.012) (0.017) (0.017) (0.016) (0.016)
Constant 2.515*** 2.456*** 2.540*** 2.485*** 2.410*** 2.347***
(0.068) (0.068) (0.090) (0.090) (0.092) (0.092)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Endogeneity test P-val 0.406 0.490 0.470 0.519 0.750 0.845

Observations 142,010 142,010 73,936 73,936 68,074 68,074

R-squared 0.47 0.47 0.44 0.44 0.52 0.51
Adj. R-squared 0.47 0.47 0.44 0.44 0.51 0.51

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate by region in the previous year was used as an instrument for the logarithm of the unemployment rate by
region at time t.
c) Apart from regional and time fixed effects we use a set of control variables such as age, gender, marital status, city size, educational attainment, tenure, occupation,
type of work, firm size, sector and industry.
d) The endogeneity test is defined as the difference of two Sargan-Hansen statistics: one for the equation with the smaller set of instruments, where the suspect regressor
is treated as endogenous, and one for the equation with the larger set of instruments, where the suspect regressors are treated as exogenous. Under the null hypothesis,
the specified endogenous regressors can actually be treated as exogenous.

higher for the entire sample and the female subsample, and slightly lower
for the male subsample. Moreover, the impact of RPP on the value of the
unemployment elasticity is exactly the same as that found in the micro-
economic specification, leading in both cases to a significant decrease.
The local unemployment elasticity drops more than 55% for the full
sample (from
0.055 to
0.024), by more than 50% in the case of males
(from
0.060 to
0.029) and over 60% in the case of females (from

0.050 to
0.019).
The corresponding results of the two-step approach for Polish NUTS2
regions are shown in Table 5. The application of RPP leads to a decrease
of more than 20% in the unemployment elasticity for the full sample
(from
0.057 to
0.049), around 15% for males (from
0.084 to

0.071) and just under 30% for females (from
0.045 to
0.032). Note
that in the case of both the US states and the Polish NUTS2 regions, the
unemployment elasticity becomes statistically insignificant when
regionally adjusted wages are used instead of nationally adjusted wages.
These findings seem to confirm that the inclusion of regional price
indices may indeed significantly influence the results of local wage curve
analyses. In fact, the unemployment elasticities estimated for regionally
adjusted wages barely exceed
0.05 in all the cases analyzed and are
significantly lower in absolute terms than the
0.1 value found in the
wage curve literature (Blanchflower and Oswald, 2005). Nonetheless, the
wage curve relationship exists in all the specifications tested (with the
exception of the female subsample in the two-step approach). We believe
that the larger drop of unemployment elasticity for the US, as compared
to Poland, can be a reflection of more local bargaining in the US.
We now turn to the spatial wage curve in order to test the hypothesis
concerning the possible significant overestimation of spatial spillovers.
Table 6 shows the results of the analysis of the spatial wage curve for the
US states. It is based on the contiguity weights matrix and the same

7
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Table 4
The US unemployment elasticity of hourly wages – traditional wage curve in the two-step approach.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.055***
0.024*
0.060***
0.029*
0.050***
0.019

(0.016) (0.013) (0.019) (0.016) (0.017) (0.015)
Constant 1.633*** 1.615*** 1.496*** 1.478*** 1.715*** 1.697***
(0.028) (0.022) (0.032) (0.026) (0.029) (0.025)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 392 392 392 392 392 392

R-squared 0.97 0.95 0.96 0.95 0.98 0.95
Adj. R-squared 0.97 0.95 0.96 0.95 0.98 0.95

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate by state in the previous year was used as an instrument for the logarithm of the unemployment rate by
state at time t.
c) Second-step estimates. Individual level variables (age, gender, race, marital status, educational attainment, occupation and industry) were included in the first step.

Table 5
The Polish unemployment elasticity of hourly wages – local wage curve in the two-step approach.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.066***
0.052***
0.084***
0.071***
0.045**
0.032

(0.015) (0.017) (0.021) (0.023) (0.020) (0.022)
Constant 2.825*** 2.665*** 2.841*** 2.777*** 2.510*** 2.455***
(0.045) (0.051) (0.063) (0.067) (0.059) (0.064)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 256 256 256 256 256 256

R-squared 0.99 0.99 0.99 0.99 0.99 0.99
Adj. R-squared 0.99 0.99 0.99 0.99 0.99 0.99

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate by region in the previous year was used as an instrument for the logarithm of the unemployment rate by
region at time t.
c) Second-step estimates. Individual level variables (age, gender, race, marital status, educational attainment, occupation and industry) were included in the first step.

Table 6
The US spatial unemployment elasticity of hourly wages by worker type – the impact of PPP deflators on the spatial wage curve.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.036***
0.013***
0.047***
0.023***
0.029***
0.006

(0.005) (0.005) (0.007) (0.007) (0.007) (0.007)
Spatial spillovers
0.056**
0.044***
0.065***
0.054***
0.045***
0.033***
(0.005) (0.005) (0.007) (0.007) (0.008) (0.008)
Constant 1.817*** 1.886*** 1.621*** 1.691*** 1.693*** 1.763***
(0.013) (0.013) (0.018) (0.018) (0.019) (0.019)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 7,245,942 7,245,942 3,667,084 3,667,084 3,578,858 3,578,858

R-squared 0.38 0.38 0.39 0.39 0.38 0.37
Adj. R-squared 0.38 0.38 0.39 0.39 0.38 0.37

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate and spatial spillovers by state in the previous year was used as an instrument for the logarithm of the
unemployment rate and spatial spillovers by state at time t.
c) Spatial spillovers based on contiguity weights matrix.
d) Apart from regional and time fixed effects we use a set of control variables such as age, gender, race, marital status, educational attainment, occupation and industry.

8
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Table 7
The Polish spatial unemployment elasticity of hourly wages by worker type – the impact of PPP deflators on the spatial wage curve.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.027
0.024
0.036
0.034
0.022
0.018

(0.019) (0.019) (0.026) (0.026) (0.027) (0.027)
Spatial spillovers
0.094***
0.063**
0.099**
0.068*
0.085**
0.055
(0.028) (0.028) (0.039) (0.039) (0.039) (0.039)
Constant 2.658*** 2.552*** 2.690*** 2.589*** 2.541*** 2.431***
(0.075) (0.075) (0.101) (0.101) (0.103) (0.103)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 142,010 142,010 73,936 73,936 68,074 68,074

R-squared 0.47 0.47 0.44 0.44 0.52 0.51
Adj. R-squared 0.47 0.47 0.44 0.44 0.51 0.51

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate and spatial spillovers by region in the previous year was used as an instrument for the logarithm of the
unemployment rate and spatial spillovers by NUTS2 region at time t.
c) Spatial spillovers based on contiguity weights matrix.
d) Apart from regional and time fixed effects we use a set of control variables such as age, gender, marital status, city size, educational attainment, tenure, occupation,
type of work, firm size, sector and industry.

estimation approach as for the local wage curve. We also tested other
weight matrices (inverse or inverse square distance). However, the re-
sults were very similar. Note that given the considerable size of US states,
the existence of spatial labor market spillovers may raise certain con-
cerns. However, as previously argued, in the presence of local monop-
sonistic competition, employers’ wage policies are likely to take into
account the average wage level in the surrounding areas (Longhi et al.,
2006). The existence of spatial interdependencies among the US states is
also found in many other papers (e.g., Baicker, 2001; Hamilton and
Owyang, 2012; Park and Hewings, 2012; Ojede et al., 2018).
In columns 1–2 of Table 6, we compare the results for the whole
sample and confirm the validity of our research hypothesis. Applying
regional price deflators leads to a reduction in the local unemployment
elasticity by
0.023 (from
0.036 to
0.013), whereas the spatial
spillover decreases by
0.012 (from
0.056 to
0.044). These results
imply that the local unemployment elasticity decreases by more than
60% and the spatial spillovers diminish by more than 20%. The impact of
regional price deflators on both the local unemployment elasticity and
the spatial spillovers seems to be similar across subsamples. The local
unemployment elasticity falls by
0.024 for males and
0.023 for
females. The decreases in the spatial spillover are
0.011 and
0.012
respectively. However, the relative magnitude of the impact is much
larger for females. Here, the local unemployment elasticity decreases by
almost 80% (and becomes statistically insignificant) and the spatial
spillover decreases by more than 25%. In the case of males, the re-
ductions are over 50% and less than 20% respectively. In all of the above
cases, the differences between the elasticities are statistically significant.
In the case of Polish regions, the local unemployment elasticity in the
full sample decreases by
0.003 and spatial spillovers decline by
0.031
(see Table 7), a reduction of around 30% in the latter case. Note that, as
expected, the spatial spillover elasticity is much higher for Polish NUTS2
regions than for the US states. For the male subsample, the local unem-
ployment rate falls by
0.002 and spatial spillovers decrease by
0.031.
For females, the reduction amounts to
0.004 and
0.030 respectively.
In similar fashion to the US states, the value of spatial spillovers is always
higher than that of the local unemployment elasticity, although in the
Polish case, the local unemployment elasticity is no longer statistically
significant. This may be due to the fact that the wage curve evolves over
time depending on labor market conditions. Hence, the local unem-
ployment elasticity may be higher when conditions are not favorable and

Table 8
The US unemployment elasticity of hourly wages – spatial wage curve in the two-step approach.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.052***
0.026**
0.053***
0.027*
0.051***
0.025*

(0.016) (0.013) (0.019) (0.015) (0.017) (0.015)
Spatial spillovers
0.006 0.005
0.015
0.004 0.003 0.014
(0.010) (0.008) (0.012) (0.010) (0.011) (0.009)
Constant 1.637*** 1.611*** 1.507*** 1.481*** 1.713*** 1.687***
(0.030) (0.024) (0.034) (0.029) (0.031) (0.027)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 392 392 392 392 392 392

R-squared 0.97 0.95 0.96 0.95 0.98 0.95
Adj. R-squared 0.97 0.95 0.96 0.95 0.98 0.95

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate by state in the previous year was used as an instrument for the logarithm of the unemployment rate by
state at time t.
c) Second-step estimates. Individual level variables (age, gender, race, marital status, educational attainment, occupation and industry) were included in the first step.

9
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Table 9
The Polish unemployment elasticity of hourly wages – spatial wage curve in the two-step approach.
All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.025
0.026
0.040
0.041
0.003
0.004

(0.024) (0.026) (0.033) (0.035) (0.030) (0.033)
Spatial spillovers
0.095***
0.062*
0.102**
0.068
0.098**
0.064
(0.034) (0.037) (0.049) (0.049) (0.044) (0.049)
Constant 2.966*** 2.756*** 2.991*** 2.878*** 2.664*** 2.550***
(0.059) (0.069) (0.083) (0.088) (0.081) (0.092)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 256 256 256 256 256 256

R-squared 0.99 0.99 0.99 0.99 0.99 0.99
Adj. R-squared 0.99 0.99 0.99 0.99 0.99 0.99

Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the logarithm of the unemployment rate by region in the previous year was used as an instrument for the logarithm of the unemployment rate by
region at time t.
c) Second-step estimates. Individual level variables (age, gender, race, marital status, educational attainment, occupation and industry) were included in the first step.

lower (or even not statistically significant) when they improve. The
current situation in the Polish labor market is characterized by an his-
torically low unemployment rate that fell to less than 6% in 2019. From
an economic point of view, this may suggest that employers are more
aware of the labor market conditions in the surrounding areas than in the
local one. Earlier work by Baltagi and Rokicki (2014) shows the existence
of a statistically significant spatial wage curve in Poland over the period
1999–2010. However, also in their case, the local unemployment rate is
found to be insignificant for certain subsamples (e.g., women).
Once again, we assess the robustness of our findings by applying a
two-step approach. Table 8 reveals the results of second-step estimates
for the US states. This time we find that the impact of RPP on local un-
employment rates is very similar for both the full sample and the sub-
samples. The reduction in the unemployment elasticity is close to 50% in
all of the cases. At the same time, the spatial spillover coefficient is very
small and not statistically significant. In this sense, the results of the two-
step approach differ from those based on the Mincerian approach and
may suggest that due to the size of US states employers are less aware of
labor market conditions in the surrounding areas. Another possible
explanation is the heterogeneity of spillover effect. For instance, Illinois
firms are likely to be aware of wage rates in Wisconsin and Indiana. At
the same time, however, there is much less spillover among states that
belong to Great Plains and the ones on the western coast.
In the case of Polish NUTS2 regions, the results of the two-step
approach are almost identical to those obtained with the one-step
approach, in terms of both the size of the coefficients and their statisti-
cal significance (see Table 9). The local unemployment rate is not sta-
tistically significant in any of the specifications. This again may indicate
that employers are equally or more aware of the labor market conditions
in the surrounding areas than in the local one. Such a result may seem
reasonable in the Polish case where the regions are much smaller as
compared to the US states. On the other hand, in the case of the full
sample, the spatial spillover is statistically significant and decreases by
just over 30% when RPP is applied. The reduction in the spatial spillovers
in the case of the subsamples is similar in terms of size, but with a loss of
statistical significance.
These results seem to confirm our hypothesis concerning a possible
overestimation of the wage curve. We show that both the local unem-
ployment elasticity and the magnitude of the spatial spillovers decrease
significantly when regional price differentials are taken into account,
though the magnitude of the impact may differ for both countries. This,
in turn, depends on the strength of the relationship between regional
price indices and wages and may be related to differences in the size of
the corresponding regions. We also find that the impact may differ in the
male and the female subsamples.
6. Conclusions
This paper reconsiders the negative relationship between wages and
the regional unemployment rate, commonly referred to as the wage
curve. We apply regional price deflators in order to adjust regional wages
and estimate the unemployment elasticity using microdata for the US
states and Polish NUTS2-level regions. To test the impact of regional
prices on the wage curve, we apply a variety of approaches, including a
Mincerian one, a spatial spillovers approach and the two-step approach
developed by Bell et al. (2002). Our findings indicate that previous re-
sults of wage curve studies overestimated the negative relationship be-
tween (real) wages and unemployment rates, which may imply that the
unemployment elasticity is significantly lower than the commonly
adopted
0.1. This is due to the negative relationship between regional
prices and unemployment. We show that this holds over a longer period
of time even though the differences between regionally adjusted and
nationally adjusted wages may evolve. Nonetheless, we confirm that the
wage curve also exists when regionally adjusted wages are considered
instead of nationally adjusted ones. In accordance with the theory of
monopsonistic competition, this inverse relationship is related to the
costs faced by both firms and workers.
We find that in the case of the local wage curve, the unemployment
elasticity estimated for the full sample falls from
0.049 to
0.023 for
the US states and from
0.067 to
0.051 for the Polish NUTS2 regions.
These results signify a reduction of more than 50% and less than 25%
respectively. We also find, however, that the impact may differ for the
male and female subsamples. For instance, in the case of the US regions,
the reduction in the unemployment elasticity is around 40% for men and
60% for women. In the case of Polish NUTS2 regions the unemployment
elasticity drops by 20% and 30% respectively. The results of the two-step
approach are in line with the one-step approach based on the Mincerian
framework. The impact of RPP on the unemployment elasticity is almost
exactly the same in terms of magnitude.
Regional price deflators are also found to have sizable impact in the

10
B. Rokicki et al. Economic Modelling 102 (2021) 105582

case of the spatial wage curve. Here, spatial spillovers decrease signifi-
cantly in the case of the US states and Polish NUTS2-level regions, for
both the full sample and the subsamples. The magnitude of the impact is
very similar – the results for the full sample indicate a 20% and 30%
reduction of spatial spillovers respectively. Once again, the two-step
approach estimates fully confirm our findings.
These results are important for the discussion of wage curves in
general. From the analyses presented, the expectation can be derived that
a wage curve might also exist in countries for which no regional price
indexes are available. However, the coefficient of the unemployment rate
is expected to be smaller in absolute terms than would be the case with an
estimate based on nationally adjusted wages only. Given the confirmed
inverse relationship between wages and unemployment, disadvantaged
regions are hit twice: they have relatively high unemployment rates and -
as a consequence - relatively low wages.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Acknowledgements
The authors gratefully acknowledge helpful comments received from
the editor and two anonymous referees of this journal. We would also like
to thank Ludwig von Auer (Trier University), Bernd Fitzenberger (Frie-
drich-Alexander-Universit€at Erlangen-Nürnberg), Georg Hirte (Technical
University of Dresden), Mark Partridge (Ohio State University), Jacques
Poot (University of Waikato), Raul Ramos (University of Barcelona), Karl
Taylor (University of Sheffield) and Katja Wolf (IAB) for their comments
and suggestions. We are also grateful for the feedback we received from
participants at the following meetings: Conference of the British and Irish
Section of the ERSA, Cambridge 2019; Workshop on the Wage Curve,
Schwerin 2019.
B. Rokicki gratefully acknowledges financial support from the Polish
National Science Centre under the grant UMO-2016/23/B/HS4/02370.

Statistical Appendix

Table A1
The US unemployment elasticity of hourly wages – the impact of PPP deflators on the local wage curve (Bartik measure as an instrument for the unemployment rate)

All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.111***
0.099***
0.119***
0.108***
0.099***
0.087***

(0.013) (0.013) (0.019) (0.019) (0.018) (0.018)
Constant 1.853*** 1.964*** 1.637*** 1.749*** 1.741*** 1.850***
(0.025) (0.025) (0.035) (0.035) (0.034) (0.034)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 7,245,942 7,245,942 3,667,084 3,667,084 3,578,858 3,578,858

R-squared 0.38 0.38 0.39 0.39 0.38 0.37
Adj. R-squared 0.38 0.38 0.39 0.39 0.38 0.37
Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) FE-2SLS estimation, the Bartik (1991) measure by state at time t was used as an instrument for the logarithm of the unemployment rate by state at time t.
c) Apart from regional and time fixed effects we use a set of control variables such as age, gender, race, marital status, educational attainment, occupation and industry.

Table A2
The Polish unemployment elasticity of hourly wages – the impact of PPP deflators on the local wage curve (OLS estimation)

All workers Men Women

Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted Nationally adjusted Regionally adjusted

Unemployment
0.060***
0.046***
0.069***
0.056***
0.055***
0.039***

(0.009) (0.009) (0.012) (0.012) (0.012) (0.012)
Constant 2.496*** 2.440*** 2.516*** 2.464*** 2.400*** 2.341***
(0.064) (0.064) (0.084) (0.084) (0.086) (0.086)

Region dummies Yes Yes Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Yes Yes

Observations 142,010 142,010 73,936 73,936 68,074 68,074

R-squared 0.47 0.47 0.44 0.44 0.52 0.51
Adj. R-squared 0.47 0.47 0.44 0.44 0.51 0.51
Notes.
a) Robust standard errors in parentheses; *, ** and *** represent significance at 10%, 5% and 1% levels, respectively.
b) Apart from regional and time fixed effects we use a set of control variables such as age, gender, marital status, city size, educational attainment, tenure, occupation,
type of work, firm size, sector and industry.

11
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Data Appendix

The data from the US census provide information on the state where the employee is a resident. Furthermore, the census collects data on personal
characteristics, employment spells and employers. Our data cover the 2008–2015 period and refer to the private sector only. After excluding the
unemployed, the inactive and missing observations, we are left with 7,245,942 observations.
The control variables used in the regressions are:

 Age of the individual

 Gender. Male ¼ 1 and female ¼ 2.
 Race. This variable includes 9 categories: White, Black, American Indian, Chinese, Japanese, Other Asian, Other race, Two major races, Three or
more major races.
 Marital status. This variable includes 6 categories: Married with spouse present, Married with spouse absent, Separated, Divorced, Widowed, Single.
 Education. This variable includes 11 different categories: N/A or no schooling, Nursery school to grade 4, Grade 5, 6, 7, or 8, Grade 9, Grade 10,
Grade 11, Grade 12, 1 year of college, 2 years of college, 4 years of college, 5þ years of college
 State. We distinguish all US states except Alaska and Hawaii.
 Occupation. This variable includes 9 categories: representatives of the public authorities; senior officials and managers; specialists, technicians and
other medium staff; office staff; personal services staff and dealers; farmers, gardeners, foresters and fishermen; industrial workers and craftsmen;
operators and assemblers of machinery and equipment; employed in elementary occupations.
 Industry classification. This variable defines the specific industry to which the employing establishment belongs based on the NAICS classification.
These include 17 categories: Primary Sector; Mining, Quarrying, and Oil and Gas Extraction; Utilities; Construction; Manufacturing; Wholesale
Trade; Retail Trade; Transportation and Warehousing; Information; Finance and Insurance; Real Estate and Rental and Leasing; Professional, Sci-
entific, and Management, and Administrative and Waste Management Services; Educational Services; Health Care and Social Assistance; Arts,
Entertainment, and Recreation; Accommodation and Food Services; Other Services, Except Public Administration.

Summary statistics – US data

Variable Obs Mean Std. Dev. Min Max

Region 7,245,942 28.07577 15.83203 1 56

Year 7,245,942 2011.514 2.311092 2008 2015
Gender 7,245,942 1.493912 0.499963 1 2
Age 7,245,942 41.47724 13.80433 16 74
marital status 7,245,942 2.970438 2.242502 1 6
education 7,245,942 7.275395 2.216177 0 11
Section 7,245,942 9.960566 4.329848 1 17
unemployment rate 7,245,942 13.30808 5.258594 3.0 13.8
occupation 7,245,942 5.429938 2.598780 1 9
hourly wage 7,245,942 18.77895 10.64067 4 50

The data from the Polish Labor Force Survey provides information on the region (at the NUTS2 level) where the employee is a resident and the size of
the city where he or she lives. Furthermore, the survey collects data on personal characteristics, employment spells and employers. Our data cover the
2000–2015 period. After excluding the unemployed, the inactive and missing observations, we are left with 142,381 observations.
The control variables used in the regressions are:

 Age of the individual

 Gender. Male ¼ 0 and female ¼ 1.
 Marital status. This variable includes 4 categories: Married, Single, Widow, Divorced.
 Education. This variable includes 7 different categories: higher (including second stage of tertiary education) ¼ 7; post-secondary ¼ 6; secondary
vocational ¼ 5; general secondary ¼ 4; vocational ¼ 3; primary and lower secondary ¼ 2; incomplete primary and no education ¼ 1.
 Region. We distinguish between 16 Polish NUTS2 regions.
 City size. This variable includes 8 categories: city with a population over 100,000 ¼ 7; population 50,000–99,900 ¼ 6; population 20,000–49,900 ¼
5; population 10,000–19,900 ¼ 4; population 5000–9900 ¼ 3; population 2000–4900 ¼ 2; less than 2000 ¼ 1; village ¼ 0.
 The individual's years of tenure at the firm.
 Type of work. Permanent contract ¼ 0, temporary contract ¼ 1.
 Occupation. This variable includes 9 categories: representatives of the public authorities; senior officials and managers; specialists, technicians and
other medium staff; office staff; personal services staff and dealers; farmers, gardeners, foresters and fishermen; industrial workers and craftsmen;
operators and assemblers of machinery and equipment; employed in elementary occupations.
 Industry classification. This variable defines the specific industry to which the employing establishment belongs, in accordance with the NACE rev. 1.1
classification. These include 15 categories: primary sector; mining and quarrying; manufacturing; electricity, gas and water supply; construction;
trade and repair; hotels and restaurants; transport, storage and communication; financial intermediation; real estate, renting and business activities;
public administration; education; health care and social work; other community, social and personal service activities; households.
 Firm size. This variable includes 5 categories: up to 10 employees ¼ 1; 11–19 employees ¼ 2; 20–49 employees ¼ 3; 50–100 employees ¼ 4; over 100
employees ¼ 5.
 Sector. Public ¼ 0, private ¼ 1.

12
B. Rokicki et al. Economic Modelling 102 (2021) 105582

Summary statistics – Polish data

Variable Obs Mean Std. Dev. Min Max

Region 142,381 16.91633 9.190708 2 32

city size 142,381 3.459373 2.945268 0 7
Year 142,381 2007.398 4.756559 2000 2015
Gender 142,381 0.479214 0.499569 0 1
Age 142,381 39.84136 11.17782 15 74
marital status 142,381 0.398297 0.746347 0 3
education 142,381 4.507392 1.658199 1 7
Sector 142,381 0.617547 0.485988 0 1
Section 142,381 10.10634 3.950736 2 18
firm size 142,381 3.371419 1.512571 1 5
type of work 142,381 0.241879 0.428223 0 1
Tenure 142,381 9.365962 9.605887 0 53
unemployment rate 142,381 13.30808 5.258594 5.5 26.3
occupation 142,381 3.698106 2.521869 0 8
hourly wage 142,381 38.10811 24.88527 2.083 1500

References Hamilton, J., Owyang, M., 2012. The propagation of regional recessions. Rev. Econ. Stat.
94 (4), 935–947.
Haworth, C., Rasmussen, D., 1973. Determinants of metropolitan cost of living variations.
Aten, B., D'Souza, R., 2008. Regional price parities: comparing price level differences
South. Econ. J. 40 (2), 183–192.
across geographic areas. Surv. Curr. Bus. 88 (11), 64–74.
Hogan, T., 1984. Geographical living-cost differentials in the United States: a
Aten, B., Figueroa, E., 2014. Real personal income and regional price parities for states
disaggregated view. Land Econ. 60 (1), 115–119.
and metropolitan areas, 2008–2012. Surv. Curr. Bus. 94 (6), 1–8.
Johansen, U., Egging, R., Johansen, K., 2019. Regional wage responsiveness to
Baicker, K., 2001. The spillover effects of state spending. J. Publ. Econ. 89 (2), 529–544.
unemployment — rural–urban heterogeneity and policy implications. Lab. Travail 33
Baltagi, B.H., Blien, U., Wolf, K., 2012. A dynamic spatial panel data approach to the
(3), 371–387.
German wage curve. Econ. Modell. 29 (1), 12–21.
Kosfeld, R., Eckey, H.-F., Lauridsen, J., 2008. Disparities in prices and income across
Baltagi, B.H., Rokicki, B., 2014. The spatial polish wage curve with gender effects:
German NUTS 3 regions. Appl. Econ. Q. 54 (2), 123–141.
evidence from the polish labor survey. Reg. Sci. Urban Econ. 49, 36–47.
Longhi, S., Nijkamp, P., Poot, J., 2006. Spatial heterogeneity and the wage curve
Baltagi, B.H., Rokicki, B., de Souza, K.B., 2017. The Brazilian wage curve: new evidence
revisited. J. Reg. Sci. 46 (4), 707–731.
from the National Household Survey. Empir. Econ. 53 (1), 267–286.
Longhi, S., 2012. Job competition and the wage curve. Reg. Stud. 46 (5), 611–620.
Bartik, T., 1991. Who Benefits from State and Local Economic Development Policies?
Manning, A., 2003. Monopsony in Motion. Imperfect Competition in Labour Markets.
Kalamazoo, Mich. W.E. Upjohn Institute for Employment Research.
Princeton University, Princeton.
Bell, B., Nickell, S., Quintini, G., 2002. Wage equations, wage curves and all that. Lab.
McMahon, W., 1991. Geographical cost of living differences: an update. Am. R. Estate
Econ. 9, 341–360.
Urban Econ. Assoc. J. 19 (3), 426–449.
Beraja, M., Hurst, E., Ospina, J., 2019. The aggregate implications of regional business
Nagayasu, J., 2011. Heterogeneity and convergence of regional inflation (prices).
cycles. Econometrica 87 (6), 1789–1833.
J. Macroecon. 33 (4), 711–723.
Blanchard, O., Katz, L., 1992. Regional evolutions. Brookings Pap. Econ. Activ. 1, 1–75.
Ojede, A., Atems, B., Yamarik, S., 2018. The direct and indirect (spillover) effects of
Blanchflower, D.G., Oswald, A.J., 1990. The wage curve. Scand. J. Econ. 92, 215–235.
productive government spending on state economic growth. Growth Change 49 (1),
Blanchflower, D.G., Oswald, A.J., 1994. The Wage Curve. MIT Press, Cambridge, MA.
122–141.
Blanchflower, D.G., Oswald, A.J., 1995. An introduction to the wage curve. J. Econ.
ONS, 2018. Relative Regional Consumer Price Levels of Goods and Services. Office for
Perspect. 9 (3), 153–167.
National Statistics UK, UK, 2016.
Blanchflower, D.G., Oswald, A.J., 2005. The Wage Curve Reloaded. NBER Working Paper
Park, Y., Hewings, G., 2012. Does industry mix matter in regional business cycles ? Stud.
Series, vol. 11338. Available at: SSRN: http://ssrn.com/abstract¼723307.
Reg. Sci. 42 (1), 39–60.
Blien, U., Gartner, H., Stüber, H., Wolf, K., 2009. Regional price levels and the
Ramos, R., Duque, J., Surinach, J., 2010. Is the wage curve formal or informal? Evidence
agglomeration wage differential in western Germany. Ann. Reg. Sci. 43 (1), 71–88.
for Colombia. Econ. Lett. 109, 63–65.
Blien, Uwe, Phan, thi Hong Van, 2019. Effects of regional unemployment on wages in
Reed, W., 2015. On the practice of lagging variables to avoid simultaneity. Oxf. Bull.
vietnam. In: Paper Prepared for Presentation at the North American Regional Science
Econ. Stat. 77 (6), 897–905.
Conference 2019.
Rhee, H.J., Song, J., 2020. Wage rigidities and unemployment fluctuations in a small
Borjas, G.J., 2013. Labor Economics, sixth ed. McGraw-Hill, Boston.
open economy. Econ. Modell. 88, 244–262.
Brandt, L., Holz, C., 2006. Spatial price differences in China: estimates and implications.
Rokicki, B., 2015. Regional price indices and real wage equalization in Poland. In:
Econ. Dev. Cult. Change 55 (1), 43–86.

Mussida, C., Pastore, F. (Eds.), Geographical Labor Market Imbalances. Recent
Cadil, J., Mazouch, P., Musil, M., Kramulova, J., 2014. True regional purchasing power:
Explanations and Cures. AIEL Series in Labour Economics. Springer, pp. 355–372.
evidence from the Czech Republic. Post Commun. Econ. 26 (2), 241–256.
Rokicki, B., Hewings, G.J.D., 2019. Regional price deflators in Poland – evidence from
Card, D., 1995. The wage curve: a review. J. Econ. Lit. 33, 285–299.
NUTS2 and NUTS3 regions. Spatial Econ. Anal. 14 (1), 88–105.
Cebula, R., 1980. Determinants of geographic living-cost differentials in the United States:
Roos, M.W.M., 2006. Regional price levels in Germany. Appl. Econ. 38 (13), 1553–1566.
an empirical Note. Land Econ. 56, 477–481.
Ruggles, S., Genadek, K., Goeken, R., Grover, J., Sobek, M., 2015. Integrated Public Use
Cecchetti, S., Mark, N., Sonora, R., 2002. Price index convergence among United States
Microdata Series: Version 6.0 [Machine-Readable Database]. University of
cities. Int. Econ. Rev. 43 (4), 1081–1099.
Minnesota, Minneapolis.
Chung, S., Hewings, G., 2019. A short exercise to assess the effects of temporal and spatial
Sanroma, E., Ramos, R., 2005. Further evidence on disaggregated wage curves: the case of
aggregation on the amounts of spatial spillovers. In: Franklin, R. (Ed.), Population,
Spain. Aust. J. Lab. Econ. 8 (3), 227–243.
Place, and Spatial Interaction, New Frontiers in Regional Science: Asian Perspectives,
Vermeulen, W., van Ommeren, J., 2009. Compensation of regional unemployment in
vol. 40. Springer, Singapore.
housing markets. Economica 76 (301), 71–88.
Deller, S., Shields, M., Tomberlin, D., 1996. Price differentials and trends in state income
Weinand, S., von Auer, L., 2020. Anatomy of regional price differentials: evidence from
levels: a research Note. Rev. Reg. Stud. 26 (1), 99–113.
micro price data. Spatial Econ. Anal. https://doi.org/10.1080/
Dreger, Ch, Kosfeld, R., 2010. Do regional price levels converge? Jahrb. Natl. Stat. 230
17421772.2020.1729998.
(3), 274–286.
Wozniak, A., 2010. Are college graduates more responsive to distant labor market
European Communities/OECD, 2006. EUROSTAT-OECD Methodological Manual on
opportunities? J. Hum. Resour. 45 (4), 944–970.
Purchasing Power Parities. Office for Official Publications of the European
Zeilstra, A., Elhorst, P., 2014. Integrated analysis of regional and national unemployment
Communities.
differentials in the European union. Reg. Stud. 48 (10), 1739–1755.
Fingleton, B., Palombi, S., 2013. The wage curve reconsidered: is it truly an ‘empirical law
of economics’? R egion et Developpement 38, 49–92.

13