. 7
( 11)


solve this dilemma, so the argument goes, is to accept high trade exposure
while simultaneously adopting comprehensive social programs to compen-
sate people for increased levels of risk (see also Ruggie 1983).
Figure 5.1 shows that the growth in government spending does indeed
track the expansion in trade, and the thesis is logically consistent with the
argument presented in this book. Although never linked explicitly to limited
mobility of skills, if globalization leads to greater labor market volatility, one
Forces of Change

should expect demand for social protection to rise. Globalization is, how-
ever, probably only one piece in the welfare state expansion puzzle. The
reason is that participation in international trade opens opportunities to
escape excessive dependence on small home markets and to diversify risks,
much the same way as mutual stock funds spread the risks across industries
and, increasingly, across national markets. Although trade implies greater
specialization, and hence more concentrated risks, the bulk of the trade be-
tween advanced economies is intra-industry and does not lead to excessive
dependence on one or a few industries. Indeed, many industries in coun-
tries with small home markets can prosper only by taking advantage of the
production of scale that global markets enable. As I show later in this chap-
ter, with appropriate controls, the relationship between trade and spending
depicted in Figure 5.1 is weak and sometimes disappears altogether.
As Tom Cusack and I have argued (Iversen and Cusack 2000), a more
important source of welfare expansion has been deindustrialization. As



Spending as percent of GDP

18 Government transfers


Government consumption




1955 1965 1975 1985 1995

Figure 5.1 Spending, trade, and deindustrialization in seventeen OECD coun-
tries, 1952“1995.
Notes: Government consumption is all spending on public services less military spending;
transfers are all government transfers less interest payments and subsidies; trade is exports
plus imports divided by the GDP.
Sources: OECD, National Accounts, Part II: Detailed Tables (various years).

Coping with Risk

indicated in Figure 5.1, this variable, which is explained in detail later in
this chapter, is (like trade) also closely related to the evolution of spending,
and the relationship (unlike trade) turns out to hold up with control for
a large number of potentially confounding factors. There are good rea-
sons why this would be the case. The labor market dislocations associated
with deindustrialization are considerable and compare in magnitude to the
movement of workers from the countryside to the city during the industrial
revolution. In 1960, for example, about 60 percent of the labor force in the
OECD area was employed in the primary sectors; thirty-¬ve years later this
¬gure is down to about 30 percent. This massive shift in employment is the
outgrowth of deep forces of technological change coupled with progres-
sive market saturation in manufactured goods and shifts in demand toward
services “ structural-technological conditions that also characterized the
great transformation from agriculture to industry.
Perhaps more surprisingly, there is also considerable variance in the
speed of deindustrialization across countries.1 For example, in an early
industrializing country like the United States, industrial employment as a
percentage of the adult population declined by only 3 percentage points
between 1960 and 1995, whereas for a late industrializer like Sweden, the
¬gure is 13 percent. If we add to these ¬gures the decline of agricultural
employment, the numbers increase to 6 and 22 percent, respectively. The
difference in these numbers would translate into 23 million lost jobs if the
United States had gone through the same process of deindustrialization as
Sweden did from 1960 to 1995. The reason for this difference is that the
United States industrialized earlier than Sweden, did so at a slower pace,
and never became as heavily dominated by industry as did Sweden.
But even though it is sensible to assume that shifts in the employment
structure, proxied by deindustrialization, raise the demand for social pro-
tection, the argument in this book implies that the effect is conditional on
the skill system and on the electoral system. The more an economy empha-
sizes speci¬c skills through the training system, the less portable skills will
be, and the greater the effects of shocks such as deindustrialization on de-
mand for protection. In addition, because the aggregation of preferences is
mediated by political institutions “ in particular electoral systems as argued

1 There seems to be a misconception that deindustrialization is uniform across countries, and
therefore cannot explain cross-national variance in the speed of welfare state expansion. At
least, this is one of the only reasons I can imagine for why not a single large-N, cross-national
study of the welfare state has focused on the effects of deindustrialization.

Forces of Change

in detail in Chapter 4 “ we expect the effects of shocks to be conditional
on these institutions. Electoral systems shape the capacity of governments
credibly to commit to social protection and affect the ideological hue of
the government. The magnitude of the effect of shocks should therefore
be rising in the proportionality of the electoral system.
This chapter is organized into three sections. The ¬rst elaborates on
the argument, presents some background data, and speci¬es the empirical
hypotheses to be tested. The second tests the argument using time-series
data for sixteen advanced democracies, while the ¬nal section discusses the
long-term political consequences of the fact that the occupational structure
has been gradually stabilizing in the last couple of decades.

5.1. Deindustrialization and the Role of Institutions
The importance of changes in the occupational structure depends on the
transferability of skills and social bene¬ts. A main argument of this book
is that transferable skills protect workers against market vagaries, whereas
speci¬c skills expose them to risks. Labor market risks therefore arise across
the interfaces between economic sectors requiring very different types of
skills. This logic is reinforced when we consider that privately provided
social bene¬ts, such as health insurance and pensions, also tend to be con-
strained by the transferability of skills. The reason is that when skills are
¬rm-speci¬c, employers only have an incentive to provide nontransferable
bene¬ts, both as a tool of control over the workforce and as an incentive
for their employees to acquire additional ¬rm-speci¬c skills (Mares 1998).
Correspondingly, if skills are industry-wide, there is a rationale for employ-
ers in that industry to provide bene¬ts that are transferable across ¬rms “
but only within the industry. The latter clearly depends on the ability of
employers to collude in the provision of both skills and bene¬ts. The point
is simply that the transferability of bene¬ts will never exceed the transferability
of skills in the absence of state intervention.
The approximate correspondence between the scope of employer-
sponsored insurance and the transferability of skills means that if a worker
loses his or her job and must either transgress a skill boundary or remain
nonemployed, labor market power and bene¬ts will be forfeited or down-
graded. In some cases, this implies that workers are left outside employment
with few or no means of support; in other cases, it means that workers ¬nd
new jobs at substantially reduced wages and bene¬ts. It is therefore only
through the mediation of the state that workers can protect themselves
Coping with Risk

against the risks of major shifts in the economic and occupational structure.
Such protection comes in the form of state-guaranteed health and old-age
insurance (which makes it possible to move across sectoral interfaces with-
out losing bene¬ts), as well as through early retirement and certain forms
of disability insurance that facilitate a relatively painless exit from the labor
market (and therefore make it possible not to have to move across the skill
interfaces). Even though transfer income is usually related to past earnings,
there is unambiguous evidence for all OECD countries that posttax and
transfer income is more equally distributed than the pretax and transfer
income. We already saw this in Chapter 1 (see especially Figure 1.4), and it
was further documented in the analysis of redistribution in Chapter 4. This
is even more true for government services, which are usually not income
related, although public services do not always serve protection purposes.
It is important to point out that state accommodation of demands for
protection against labor market risks is not necessarily opposed by em-
ployers, as commonly assumed in the welfare state literature. Without
assurances from the state, workers will be less inclined to make risky in-
vestments in nontransferable skills, and many employers depend on work-
ers making precisely such investments. Especially with the transition to
more knowledge-intensive forms of production, ¬rms that rely on ¬rm- and
industry-speci¬c skills share with their employees an interest in strength-
ening the aspects of the welfare state that reduce the riskiness for workers
of making investments in speci¬c skills. While this implication is clearly
at odds with the standard assumption that business always opposes social
spending, it is consistent with an emerging new body of scholarship that
documents the supportive and often proactive role of employers in develop-
ing and shaping the modern welfare state (Martin 1995, 2003; Mares 1998,
2003; Swenson 2002).
Like the distinction between agriculture and industry in the latter half of
the previous century, the distinction between manufacturing and services
represents one of the most important economic interfaces affecting the
transferability of skills in the latter half of the 20th century. Even low-skilled
blue-color workers, almost all males, ¬nd it exceedingly hard to adjust to
similarly low-skilled service sector jobs because they lack something that, for
want of a better word, may be thought of as a form of social skills. In addition,
employers in the two sectors are usually organized into different associations
and do not cooperate in the provision of training or bene¬ts. In addition, a
switch in employment across the two sectors typically requires workers to
change their membership in unions and unemployment insurance funds.
Forces of Change

But the growth of the welfare state is not an automatic response to
deindustrialization (or to other shocks such as trade for that matter). As
I showed in Part II, the translation of sectoral shifts in the economy into
social policy depends on two key institutional variables: the skill system
and the electoral system. When skills can be unemployed to different
degrees, and the risk of skills being unemployed depends on the speci-
¬city of those skills, shocks to the labor market will affect those with
speci¬c skills more than those with general skills. In turn, this translates
into different pressure on the welfare state in countries with different skill
Going back to the model of preferences presented in Chapter 3, if we
treat the long-term probability of entering the state of the world where
speci¬c skills are unemployed (State II in Figure 3.1) as a variable, an indi-
vidual™s demand for social protection is the interaction between this variable
and the speci¬city of his or her skills. Let us therefore recall the expression
for the probability of being in the “general skills only state” (State II “ i.e.,
a state of the work where speci¬c skills are worthless):
β = (1 ’ r) · e
Assume now that the (equilibrium) level of employment, e, is constant.
Then β is a positive function of (1 ’ r), which is the probability of un-
employed workers being reemployed into a state of the world where
their speci¬c skills are unutilized. The harder it is for unemployed to
¬nd jobs where their speci¬c skills are used, the higher the probability
of some day experiencing a drop in income, and the greater the demand
for income protection, R. The central claim is therefore that deindustri-
alization, and other major shocks to the labor market, have the effect of
raising (1 ’ r) and hence the demand for social protection. The mag-
nitude of the effect depends on the speci¬city of skills as illustrated in
Figure 5.2.
The effect of exogenous shocks is also likely to depend on the electoral
system. There are two arguments. The ¬rst concerns the demand for in-
surance. As argued in Chapter 4, the fundamental problem in the provision
of social insurance is that when a shock hits, those who are affected will not
likely be the ones setting policy. Those who have not been directly affected
will update their subjective assessment of risks, but they only have an in-
terest in compensatory policies if such policies can be seen as a premium
for protection against future shocks. Yet, current voters can only commit
the government for one term at a time, and there is no way to bind future
Coping with Risk

voters to the policy preferences of current voters. The time-inconsistency
problem means that shocks that would raise demand for protection will not
necessarily be translated into actual policies.
As argued in Chapter 4, there are two institutional remedies for the
time-inconsistency problem. The ¬rst is that political parties with detailed
policy programs and highly developed party organizations, especially links
to unions, limit the ability of leaders to give in to short-term electoral incen-
tives and constrain the choice set of voters to alternatives that are optimal
in the long run. However, this solution does not eliminate the temptation
for party leaders to offer tax cuts and to shun long-term investment in social
protection. And this temptation is particularly high in majoritarian systems
where the reward for winning the next election is great.
The second argument goes back to the idea that PR electoral systems give
centrist parties an incentive to ally with left parties for current redistributive
purposes. If left parties tend to represent voters who are at greater risk,
the preferences of these voters will be represented in coalition bargaining.
The same is not true in a majoritarian system where the median voter will


High s/g
support for

support for

Low s/g

Reduced Increased
risk, β
exposure exposure
Shock to the risk distribution
Figure 5.2 Support for redistribution as a function of risk.

Forces of Change

ignore the distributive preferences of those with little income. Indeed, these
preferences present a threat to the interests of the median voter, who might
become more prone to vote for the center-right as a result.
But a shock to the income distribution may also negatively affect the
median voter and hence drive up demand for redistributive spending. Still,
the difference between PR and majoritarian systems remains. To see this,
imagine a means-preserving increase in inequality. In this situation, both
the poor and the median voter will prefer more redistribution. In a PR
system, this will come about as a compromise between the center and left
parties, where those who have lost the most in the “tail” of the distri-
bution will have as much in¬‚uence as those in the center who are likely
to have lost the least. In a majoritarian system, by contrast, although the
median voter will prefer more spending, he or she will also be more fear-
ful of a center-left party where the poor set policy. The ¬rst effect will
make the median voter more likely to vote center-left, the second effect
less likely to do so. The net effect is a less signi¬cant shift in policies
than under PR as illustrated in Figure 5.3. As in the case of skill systems,


Rising Majoritarian
support for
Rising risk
support for


Declining Rising
income income
Shock to the income distribution
Figure 5.3 Support for redistribution as a function of shocks to income.

Coping with Risk

Figure 5.4 Deindustrialization and change in welfare spending for sixteen OECD
countries, 1960“93.
Notes: Government spending is the average annual change in civilian government consumption
plus social transfers as a percentage of GDP; deindustrialization is the average annual reduction
in employment in industry plus agriculture as a percentage of the working age population.
Sources: OECD, Labour Force Statistics (various years); OECD, National Accounts, Part II:
Detailed Tables (1997).

the elasticity of government spending to shocks depends on the type of
electoral system.

5.2. Empirical Evidence2
Figure 5.4 shows the association between the average annual ¬gures for
deindustrialization “ in the sense of the joint employment losses in indus-
try and agriculture “ and the comparable ¬gure for the expansion in total
government spending. As expected, there is a positive association that is of
about the same strength as the cross-national correlation between open-
ness and spending. There is some clustering around the mean, but this is
mainly the result of averaging over a thirty-¬ve-year period (as will become

2 This section builds on Iversen and Cusack (2000).

Forces of Change

apparent later). Most of the countries have gone through periods of both
relatively slow and relatively rapid deindustrialization, and the temporal
order of these swings varies. In addition, the effects of deindustrialization
are institutionally mediated, and similar magnitudes can, therefore, have
different effects in different countries. All this can be captured through
multiple regression analyses of time-series data, to which I turn next.
The evidence covers sixteen OECD countries over a thirty-¬ve-year
period from 1960 to 1995.3 This period represents the historically most
dramatic phase of growth in welfare spending with transfer payments more
than doubling from 8.5 percent of GDP in 1960 to over 20 percent in 1995,
and government consumption increasing from about 9 percent of GDP in
1960 to over 16 percent in 1995. The cross-national variance in transfers
(measured by the coef¬cient of variation) declined somewhat over time,
but it rose for government consumption. Measured in terms of the dif-
ference between the smallest and biggest “spenders,” divergence increased
in both categories of spending. More precisely, the transfers payment gap
increased from 10 to 15 percent of GDP, and the government consump-
tion gap, from 6 to 15 percent of GDP. The data thus represents not only
considerable intertemporal variance but also large cross-national differ-
ences “ a good testing ground for competing explanations of welfare state
In addition to overall spending on transfers and services, the analysis also
examines spending that is speci¬cally targeted for purposes of social protec-
tion. Speci¬cally, I use data on consolidated central government spending
on social security, health, and welfare as a percent of GDP, which the In-
ternational Monetary Fund (IMF) has compiled since 1970 for fourteen
of these sixteen cases (IMF, Government Finance Statistics Yearbook, various
years). The OECD collects similar data, but they only go back to 1980,
thereby missing a critical phase of deindustrialization and welfare state ex-
pansion. The IMF data does have one important limitation: It excludes
spending that is ¬nanced locally (as opposed to spending ¬nanced by cen-
tral government grants). Nevertheless, it captures core social expenditures
that can be expected to respond to pressures for more social protection. As
it turns out, total transfers and IMF™s measure have a correlation of .9, so
the ¬rst is a good proxy for the latter.

3 The countries are Australia, Austria, Belgium, Canada, Denmark, Finland, France, West
Germany, Italy, Japan, Netherlands, Norway, Sweden, Switzerland, United Kingdom, and
United States.

Coping with Risk

5.2.1. Model Speci¬cation
The empirical analysis falls into two parts. In the ¬rst, I focus on the effects
of deindustrialization, ignoring interactive institutional effects. This allows
for a fairly detailed analysis of the intertemporal properties of the effects
of deindustrialization and other potential explanatory variables using an
error correction model. The second part focuses on the effects of training
systems and electoral systems, using a methodology developed by Blanchard
and Wolfers (2000).
The error correction model uses changes in government spending
(de¬ned below) as the dependent variables and has the following form:
j j j
Yi,t = ±i + β1 · Yi,t’1 + β j · Xi,t’1 + β· Xi,t + (5.1)

where Y is a spending variable (transfers, consumption, or social spending),
and X is an independent variable. The subscripts i and t index countries
and time periods, respectively, while the superscript j indexes a particular
independent variable. is the ¬rst difference operator. Note that the model
uses country-speci¬c intercepts (“country dummies”).
This model has a number of useful properties (see Beck and Katz 1996).
First, the parameter for the lagged dependent level variable, β1 , provides
an easy check on equilibrium properties. β1 should be between ’1 and 0 to
ensure that the incremental effects of a shock to any exogenous variable are
progressively reduced over time, causing spending to converge to a long-
term equilibrium. For readers more familiar with models that use the level of
spending on the left-hand side, the current model can be reformulated into
such a model by simply adding Yi,t’1 on both sides of the equal sign. This
yields Yt,i = ± + (1 + β1 ) · Yi,t’1 + · · · , where (1 + β1 ) is the new parameter
for the lagged dependent level variable. There is a small advantage for using
¬rst differences, however, because the model yields estimates of R2 that are
more informative of the variance explained by the independent variables of
The other useful feature of the present model is that it enables us to sep-
arate out the permanent and transitory effects of any independent variable.
Although not intuitively obvious, it can be shown that the parameter for a
lagged independent level variable, Xt’1 , is a measure of the permanent (or
lasting) effect of a one-off change in that variable, while the parameter for

4 The reason for the advantage is that in the model using levels of spending on the left-hand
side, much of the variance will be accounted for by the lagged dependent variable, showing
simply that current spending depends on past spending.

Forces of Change

a change variable, Xt , is a measure of the transitory (or passing) effect of a
one-off change in that variable (see Beck 1992). The long-term permanent
effect of an independent variable can be calculated by dividing the parameter
for the lagged level of that variable by minus the parameter for the lagged
dependent level variable: β j / ’ β1 (assuming that β1 between 0 and ’1). If
a variable exhibits only transitory effects (i.e., if only the parameter for its
¬rst difference is different from zero), spending will eventually revert back
to its original level unless the independent variable changes continuously
(assuming again that β1 is between 0 and ’1). Because all the theoretical
variables are de¬ned as proportions (either of GDP or of the working age
population), they cannot grow (or fall) inde¬nitely and will therefore have
no lasting effect on spending unless the parameters for the their lagged lev-
els are signi¬cant.5 In the interactive institutional model presented later in
this chapter, I focus on levels, and hence permanent effects only (basically
to keep the model manageable).
As noted, all regressions were run with a full set of country dummies (or
country-speci¬c intercepts) to control for nationally speci¬c effects. For ex-
ample, Esping-Andersen (1990) has forcefully argued that the institutional
blueprints for many of today™s welfare states were established in the
pre“World War II period, and that these institutional characteristics keep
reproducing the contemporary development of the welfare state. Likewise,
the argument by Huber et al. (1993) that the greater the opportunities for
minorities to block new spending bills (i.e., the greater the number of veto
points in the political system), the less likely it is that new legislation will be
passed or implemented. Since their index of government structures varies
across my sixteen cases, but not across time, the effect will be picked up by
the country dummies.
This estimation strategy, however, leaves out one question of consider-
able theoretical interest, namely whether countries with different institu-
tions respond differently to exogenous shocks. Again, the two institutional
features that I have emphasized are the skill system and the electoral sys-
tem. The problem with including these variables in the analysis is that
they exhibit little meaningful variance across time and, therefore, are per-
fectly (or nearly perfectly) collinear with the country dummies. To over-
come this problem, I subsequently adopt an estimation strategy developed
by Blanchard and Wolfers (2000) that involves entering the institutional

5 This does not have to be the case. One of the control variables, unexpected GDP growth,
can in principle rise inde¬nitely.

Coping with Risk

variables only as interactions with the shocks that they are supposed to
condition the affect of.6
Blanchard and Wolfers propose two versions of the model, and I esti-
mate both. The ¬rst assumes that countries are exposed to uniform, and
unobservable, exogenous shocks. Because the nature of the shocks are left
entirely unspeci¬ed, the purpose is simply to determine whether countries
with different institutions respond differently to them. The shocks are prox-
ied by a set of year dummies, Dt , that are interacted with the institutional
variables, Ii :
Yi,t = ±i + Dt · (1 + β · Ii ) + β j · Xi,t + µi,t (5.2)
The common unobserved shocks in this formulation are captured by the
time dummies, and the institutional effect is captured by the parameter β.
If β is zero, it means that the effects of the shocks are identical across
institutional con¬gurations. If it is positive, it means that the relevant in-
stitutional feature (skill speci¬city or PR in our case) magni¬es the effect
of the common shocks.
The second formulation identi¬es the nature of the shock and allows it to
vary across countries. Our shock variable, of course, is deindustrialization,
Ei,t , which is simply substituted in for the time dummies:
Yi,t = ±i + Ei,t · (1 + β · Ii ) + β j · Xi,,t + µi,t (5.3)

Explanatory Variables Deindustrialization is de¬ned as 100 minus the
sum of manufacturing and agricultural employment as a percentage of the
working-age population. The base of 100 is arbitrary. For example, one
could have used the peak of employment in agriculture and manufacturing
as the base instead, and this is a number that varies across countries. It is in
fact not easy to theoretically justify the use of a particular base, but it turns
out not to be necessary. The reasons is that the statistical analysis includes a
full set of country dummies.7 As a result, if each country has a unique base,
it simply alters the nationally speci¬c intercepts, and the dummies permit
these to take on any value.
Again, the institutional variables are skill speci¬city and electoral sys-
tem. For the latter, I use the simple classi¬cation of electoral systems into

6 The dependent variable in Blanchard and Wolfer™s analysis is unemployment.
7 An F-test indicates that the country dummies belong in the model.

Forces of Change

majoritarian (0) and PR (1) that was used in Chapter 4. For the former, I use
the share of an age cohort going through a vocational training. This mea-
sure, which starts in 1980s, exhibits little meaningful variation over time
and is treated as an institutional variable. As such, it was simply extrapolated
back in time.
In addition to these variables, the analysis contains the following set of

Government center of gravity. This variable is the same as that used in
Chapter 4 and is based on three expert surveys of the left“right position
of parties (weighted by the share of parties™ seats in government).
Source: Cusack and Fuchs (2002).
Trade openness. This variable includes total exports and imports of goods
and services as a percentage of GDP. Source: OECD, National
Accounts, Part II: Detailed Tables (various years).
Capital market openness. The index is taken from Quinn and Inclan (1997)
and measures the extent to which capital markets are liberalized.
Electoral participation. The measure is voter turnout rates as recorded on
an annual basis in Mackie and Rose (1991), European Journal of Politi-
cal Research, and International Institute for Democracy and Electoral
Assistance (1997).
Unexpected growth. This is a variable emphasized in Roubini and Sachs
(1989) and is de¬ned as per capita real GDP growth at time t minus
average real per capita growth in the preceding three years. It is in-
tended to capture the logic that budgeting relies on GDP forecasts
based on performance in the recent past. If growth is unexpectedly
high, it reduces spending as a proportion of GDP. Sources: Cusack
(1991) and OECD, National Accounts, Part II : Detailed Tables (various

Two additional variables deserve particular emphasis. Both are designed to
remove nondiscretionary elements of government spending, and they are
essential to get a well-speci¬ed model.8 The ¬rst relates transfer payments
to changes in unemployment and demographics that generate “automatic”
disbursements of payments according to rules that cannot be readily al-
tered in the short run. It is standard pratice to control for such effects by

8 In my view, too little care is often taken in controlling for nondiscretionary spending, which
creates potential problems of omitted variable bias. See Iversen and Cusack (2000) for an

Coping with Risk

including variables for unemployment and the number of people above
the pension age; however, Cusack (1997) has developed a more satisfac-
tory measure that takes account of the fact that the generosity of transfers
varies across countries. The measure is referred to as automatic transfers and
de¬ned as

Automatic transfers
Unemployed + Population > 65
= Generosity (t ’ 1) · (t)

where generosity is the percentage share of transfers in GDP relative to
the percentage share of the dependent population in the total population
at time t ’ 1. In other words, changes in size of the dependent popula-
tion “ the unemployed and the retired “ causes an automatic increase in
transfers at time t, the size of which depends on the generosity of trans-
fers in the previous period (according to prevailing rules). The source for
the unemployment and population ¬gures is OECD, Labour Force Statistics
(various years).
Another nondiscretionary element of spending concerns government
consumption. Because productivity increases in public services are gener-
ally lower than in the rest of the economy, while wage and other costs
tend to follow productivity increases in the rest of the economy, the price
level of government services will “automatically” grow at a faster rate
than the general price level (the “Baumol effect”). Even at constant pro-
vision levels, government consumption will therefore increase as a share
of GDP. This nondiscretionary effect can be removed by another measure
developed by Cusack (1997). It is here called automatic consumption and is
de¬ned as

Gov consumption
Automatic consumption = (t ’ 1)
Gov de¬‚ator (t) GDP def
Gov de¬‚ator (t ’ 1) GDP def

where Gov de¬‚ator is the price de¬‚ator for government services, and GDP
de¬‚ator is the price de¬‚ator for the whole GDP. The equation simply says
that if prices on government services grow faster than the general price
level, government consumption will automatically increase by a propor-
tional amount.
Forces of Change

5.2.2. Findings Effects of Deindustrialization The results for each category
of spending are presented in Table 5.1.9 First note the effects of
deindustrialization. For each percent decline in employment in the tra-
ditional sectors, the long-term equilibrium for social transfer spending
increases by approximately 0.65 percent. The corresponding effect for gov-
ernment consumption is somewhat smaller, 0.55 percent, while the short-
term impact is to elevate the actual spending level by 1 percent for every
percent decrease in employment in the traditional sectors. The effect on
equilibrium spending on healthcare, social security, and welfare is .40 per-
cent. In other words, a standard deviation change in deindustrialization
is associated with at roughly 0.5 standard deviation change in spending,
which implies that about half of the variance in spending is explained by
the deindustrialization variable. All effects of deindustrialization are statis-
tically signi¬cant at a .01 level or better.
It is also noteworthy that the effect of deindustrialization persists over
time. Apparently spending gets “locked in” by organizational and insti-
tutional factors that are exogenous to the model. As argued by Pierson,
spending itself creates political clienteles that will press for further spend-
ing and resist attempts at retrenchment (Pierson 1994; 1996). Hence, even
though the process of deindustrialization is the causal agent in the expan-
sion of the welfare state, the disappearance of this causal agent will not
necessarily lead to retrenchment “ it will “merely” retard further expan-
sion. However, the character of the political game over welfare policies
is likely to change when compromises involving overall expansion are no
longer feasible; this conjecture deserves closer attention considering that
the process of deindustrialization is coming to a halt in many countries.
Not surprisingly, the automatic transfers variable also has a strong effect.
A parameter of .87 means that a 1 percent increase in the dependent popu-
lation is turned into additional spending amounting to .87 percent of GDP.
There is a similar effect on government consumption of changes in relative
prices (automatic consumption). Roughly speaking, if wages and productiv-
ity in the private sector rise by 1 percent, wages in the public sector are also
expected to rise by .89 percent even when productivity remains constant.

9 Tests for heteroskedasticity in both pooled regressions suggested the need to correct for
this problem, so I employed Beck and Katz™s (1995) method for deriving panel-corrected
standard errors. Separate runs using robust regression techniques (not shown) yield almost
identical results, so the ¬ndings are not driven by outliers.

Coping with Risk

Table 5.1. Regression Results for Government Spending (t-scores in parentheses)

Transfers Consumption Social Spending
’0.071——— ’0.049——— ’0.186———
Lagged dependent level
(0.022) (0.015) (0.036)
0.046——— 0.027——— 0.072———
(0.015) (0.010) (0.026)
0.120——— 0.081——— 0.096*
(0.033) (0.027) (0.053)
0.297— ’0.140
Partisanshipt’1 0.159
(0.159) (0.134) (0.196)
’0.201 ’0.020 ’0.106
(0.200) (0.147) (0.288)
Turnoutt’1 0.009
(0.006) (0.006) (0.009)
’0.005 ’0.000
Turnoutt 0.002
(0.007) (0.009) (0.010)
’0.005 ’0.012
Trade opennesst’1
(0.005) (0.002) (0.006)
0.018— ’0.006 ’0.022
Trade opennesst
(0.010) (0.005) (0.011)
Capital opennesst’1 0.007 0.022
(0.028) (0.018) (0.055)
’0.078— ’0.038
Capital opennesst 0.045
(0.053) (0.037) (0.102)
’0.077——— ’0.092——— ’0.091———
Unexpected growtht
(0.011) (0.007) (0.018)
0.865——— 0.596———

Automatic transferst
(0.097) (0.086)
’ ’
Automatic consumptiont
Adjusted R-squared 0.48 0.56 0.49
Number of observations 508 508 267
Signi¬cance levels: — <0.10; —— <0.05; ——— <0.01.
Note: The results for country dummies are not shown.

No other variable appears to have a strong and statistically signi¬cant
effect. Most surprisingly, perhaps, is the fact that none of the globalization
measures have much of an impact. There are no statistically signi¬cant ef-
fects on government transfers, and the small effect of trade openness on
government consumption has the “wrong” sign. It is conceivable that this
re¬‚ects a differential welfare effect of trade. Thus, while growing exposure
to competition from low-wage countries raises the uncertainty for those
Forces of Change

already at high risk (Wood 1994; Leamer 1996), trade may well be welfare
improving for all others (Rodrik 1997, Chapter 4). Whatever the explana-
tion, the magnitude of the effect is small and only borderline signi¬cant.
For each percentage point increase in openness, the long-term equilibrium
level of civilian government consumption declines by .07 percent. All other
effects of globalization are transitory in nature.
The absence of strong effects of government partisanship is also note-
worthy. Indeed, the one effect that is borderline signi¬cant (for transfers) is
opposite of the expectation (right-of-center governments spend marginally
more than left-of-center governments). But we must be careful in how we
interpret this ¬nding. Because the dependent variable is ¬rst differences,
and because the model includes a full set of country dummies, the variance
to be explained is entirely intertemporal. Some of the cross-country differ-
ences may be attributable to the effect of past partisan policies. Indeed, if the
country dummies are left out of the model, the positive effect right parti-
sanship has on transfers disappears, and the negative effect on consumption
becomes strong and signi¬cant.10 Both effects are further magni¬ed if we
exclude the variable automatic consumption, which depends on the extent
to which public and private sector wages are coordinated “ an attribute we
normally associate with solidaristic wage policies of the left.
Most importantly, the additive nature of the model does not allow us
to answer a key dynamic question about partisanship: Do left-leaning gov-
ernments expand social protection more than right-leaning governments
in response to exogenous shocks such as deindustrialization? This question
cannot be answered by looking only at the direct effects of partisanship.
We need an interactive speci¬cation that is outlined in Section and
is further developed in the next chapter.
More generally, the results imply that deindustrialization has an aug-
menting effect on spending, but they cannot help us answer questions
about whether the effects of deindustrialization, and potentially other forces
of change, are mediated by nationally speci¬c political-economic institu-
tions. In addition to the role of partisanship, we have theoretical reasons
to believe that the nature of both the skill system and the electoral system

10 It is also problematic to remove the dummies because it is very easy to run into omitted
variable bias. Here one also cannot do it because omitting the dummies will potentially
render the deindustrialization variable meaningless (the problem of having an arbitrary

Coping with Risk

affect national responses to deindustrialization. To explore this issue, the
next section borrows the Blanchard-Wolfers approach and uses national
institutions as econometric “prisms” for the effects of exogenous shocks. Potential Objections Before turning to the institutional story, it
is perhaps helpful to address a couple of common objections to the dein-
dustrialization argument that the reader may also harbor. The ¬rst is that
the effect of deindustrialization re¬‚ects some sort of accounting relation-
ship. The argument is that increasing public consumption means employ-
ing more people in the public sector, and because public employment is an
element in the denominator of the deindustrialization variable, there must
necessarily be a relationship between spending and deindustrialization. This
is simply wrong. The denominator is de¬ned as the total working-age pop-
ulation, not the labor force; therefore, it cannot be affected by spending
except through the birth rate. So although it is true that public employ-
ment affects an element in the denominator, it is not true that this affects
the total size of the denominator. The only logically valid argument about
reversed causality is that government spending causes employment in in-
dustry and agriculture to decline “ a possibility that will be considered in
Section 5.3.
Another objection is that as long as the natural movement of people into
retirement “ “natural attrition” “ is suf¬ciently large, deindustrialization
need not have any effect on the level of labor market risks. Yet, this argument
confuses the net effect of a set of variables with the independent effect of
these variables. Natural attrition in any market segment will increase the
job security of workers by reducing supply relative to demand, just as new
entry into a segment will increase job insecurity by raising supply relative
to demand. This is why early retirement, as a public policy, is a way to
ameliorate such insecurities, and this is why spending on early retirement
schemes is causally linked to deindustrialization. But this does not alter the
thesis that reduction in the labor force as a result of deindustrialization has
an the independent, risk-augmenting effect on the labor market. Regardless
of the level of natural attrition, we always expect deindustrialization to
increase job insecurity. Institutional Effects Table 5.2 shows the results of estimating
Equation (5.2) using nonlinear least squares regression. The dependent
variables are levels of spending; the time dummies serve as proxies for the

Table 5.2. Common Shocks, National Institutions, and Government Spending (standard errors in parentheses)

Transfers Consumption Transfers Consumption Transfers Consumption
Time effect 7.20——— 7.844——— 7.67——— 7.92——— 7.55——— 7.93———
(0.86) (0.604) (0.87) (0.61) (0.86) (0.61)
Vocational training— 0.013——— 0.004——— 0.006——— 0.001
’ ’
time dummies (0.002) (0.001) (0.002) (0.002)
PR— time dummies 0.53——— 0.18——— 0.36——— 0.15———
’ ’
(0.08) (0.05) (0.10) (0.06)
Partisanship 0.93—— 0.086 0.81—— 0.04 0.81—— 0.04
(0.40) (0.345) (0.40) (0.34) (0.39) (0.35)
0.54——— 0.58——— 0.56———
Dependency ratiot ’ ’ ’
(0.07) (0.07) (0.07)
Relative pricest ’ ’1.509— ’ ’1.53— ’ ’1.56
(0.780) (0.78) (0.78)
Minimum 5.17 7.19 5.16 7.04 4.85 6.98
Maximum 11.72 9.29 9.20 8.46 10.85 8.87
Effect 6.56 2.09 4.04 1.46 6.00 1.89
Adjusted R-squared 0.92 0.9 0.92 0.91 0.92 0.9
Number of observations 564 564 564 564 564 564
Signi¬cance levels: — < 0.10;—— < 0.05; ——— < 0.01.
Notes: The results for the interactive terms correspond to β in the statistical model. The results for country and time
dummies are not shown.
Coping with Risk

exogenous shocks. Because small changes in the balance between local and
central government spending can signi¬cantly affect the estimated effects
of period by period shocks, Table 5.2 only reports results for total govern-
ment transfers and consumption.11 To re¬‚ect that the dependent variable
refers to levels, the two automatic spending variables used earlier have also
been rede¬ned as levels. One is the ratio of the dependent population (un-
employed and those over 65) to the working-age population; the other is
the price index for public services divided by the GDP price index. Several
of the variables that either did not produce notable effects in the previous
regressions or referred to change were omitted. Most of these, globaliza-
tion in particular, can also be seen as elements in the exogenous shocks
that produce the ¬scal responses and, therefore, should not be included in
principle. As noted previously, the time dummies are deliberately designed
not to identify the nature of the shocks.
The key issue, of course, is whether governments in countries with strong
vocational training systems, or with PR electoral systems, react differently
to shocks than governments in countries with weak vocational training
systems and majoritaritarian institutions. The parameter β on the interac-
tion term provides the answer. If it is positive, it means that shocks cause
spending to increase more in countries with high values on the institutional
variable. With this in mind, it is easy to see that countries with strong vo-
cational training institutions and PR electoral systems responded to shocks
by increasing spending by a greater amount than in countries with weak
vocational training institutions and majoritarian electoral systems. The ef-
fects are positive across all spending categories and statistically signi¬cant
at a .01 level in ¬ve of the six regressions.
To gauge the substantive impact of institutions, I again follow Blanchard
and Wolfers™s methodology. The institutional variables have been de¬ned
as deviations from their cross-country means so that the effects of the time
dummies refer to a country with average values on the institutional (and
control) variables. In this way, the total time effect, shown in the ¬rst line of
Table 5.2, is the difference between the parameter on the 1995 time dummy
and parameter on the 1960 time dummy. By adding to and subtracting from
the time effect, the product between this effect and β times the minimum
and maximum values on the institutional variables, respectively, we can
differentiate the spending effects of the shocks in countries with extreme

11 The results for the IMF variable are unstable and sometimes insigni¬cant. The signi¬cant
results are consistent with those reported in this section.

Forces of Change

values on the institutional variables. For example, the ¬rst column shows
that the effect on transfers of the exogenous shocks occurring between 1960
and 1995 has been to raise spending as a percentage of GDP by 5 percent in a
country with the weakest vocational training system but by nearly 12 percent
in a country with the strongest vocational training system. These numbers
are referred to as the minimum and the maximum at the end of the table,
and what is referred to as the effect is the difference between the two. The
effect can be read as a summary measure of the impact of an institution on
any particular spending variable.
Again, both intensive vocational training and PR are associated with no-
tably higher levels of spending. This is particularly true of transfers where
the institutional effect, when measured separately, is about 5 percent of
GDP. For government consumption, it is smaller, between 1 and 2 percent.
The effects of institutions are more impressive when they are used jointly to
account for differences in the effects of shocks (the last two columns). Even
though there is quite strong collinearity between the variables (r = .7), all
parameters are in the right direction and statistically signi¬cant except for
vocational training in the case of government consumption. When entered
together, PR clearly does most of the explanatory work for consumption.
The likely explanation is that public service provision often does not serve
a clear insurance function. Daycare services and elderly care are examples.
And we know from the theoretical model that skills only matter for in-
surance preferences. Still, public employment often serves an employment
protection function, and public health care and a range of other free or low-
cost services can be seen as a form of income protection. Be that as it may,
the combined effects are large and clearly support the conclusion that in-
stitutions powerfully shape countries™ responses to exogenous shocks. And
they do so in a way that is anticipated by the institutional argument.
Did governments respond differently to the changing economic envi-
ronment during the 1980s and 1990s than they did during the 1960s and
1970s? The question is dif¬cult to answer with precision because govern-
ment spending did not change very much in the second period, leaving very
little variance to be explained. In itself, this suggests that governments ei-
ther became more constrained or reached an equilibrium level of spending
by the early 1980s where shocks were adequately addressed through au-
tomatic disbursements of transfers. Either way, it has the implication that
small measurement errors can have big effects on the results. With that
caveat, Table 5.3 reports the results by period (omitting the controls for
presentational economy).
Coping with Risk

Table 5.3. Shocks and Spending in Two Subperiods

Transfers Consumption
0.015——— 0.004—
1960“79 Vocational training
7.45——— 6.89———
Time effect
0.364——— 0.221———
6.76——— 6.96———
Time effect
1980“95 Vocational training 0.009
0.70—— ’0.02
Time effect
PR 0.239
1.45——— ’0.16
Time effect
Signi¬cance levels: — < 0.10; —— < 0.05; ——— < 0.01.

The high stability of government consumption in the second period
makes it dif¬cult to say anything with con¬dence about changes in the
effects of institutions on this category of spending. But looking at the pa-
rameters for the institutional variables across spending categories gives no
indication that the distinctiveness of government responses across insti-
tutional systems has diminished. In fact, all the parameters are larger in
the second period. Bear in mind, however, that because the time effect is
dramatically smaller in the second period (in the case of government con-
sumption it is slightly negative), the overall effect is small. Coupled with the
measurement error issue, one is well advised not to draw strong conclusions
about changes in the relative responsiveness of governments over time. But
certainly there is no evidence of convergence.
What the results presented thus far cannot tell us is anything about the
nature of the shocks. This is a virtue in the sense that the results do not
depend on any particular conceptualization of the forces of change. On the
other hand, we do care about the identity of these forces. Also, because
the model treats shocks as common, we do not allow for the possibility
that some countries have been more exposed to shocks than others. If the
extent of shocks is correlated with the institutions, the shocks rather than
the institutions could explain the divergence in policies.
Our shock variable is, of course, deindustrialization. To examine the ef-
fects of this variable, I substitute it for the time dummies in the estimating
equation (5.3). The results are shown in Table 5.4. The presentation is sim-
ilar to that in Table 5.3, but the minimum and maximum (and the effect)
are now referring to the effects of the average deindustrialization that oc-
curred between 1960 and 1995 (about 18 percent) in countries with extreme
Table 5.4. Deindustrialization, National Institutions, and Government Spending (standard errors in parentheses)

Transfers Consumption Transfers Consumption Transfers Consumption
Deindustrialization effect 4.48——— 6.95——— 4.14——— 6.67——— 4.04——— 6.74———
(0.51) (0.38) (0.51) (0.37) (0.51) (0.38)
Vocational training— 0.002——— 0.002—
’0.000 ’ ’ ’0.000
deindustrialization (0.001) (0.000) (0.001) (0.001)
PR— deindustrialization 0.64——— 0.21——— 0.62——— 0.22———
’ ’
(0.16) (0.08) (0.16) (0.08)
Partisanship 0.39 0.24 0.21
’0.77— ’0.89—— ’0.88
(0.41) (0.40) (0.40) (0.40) (0.40) (0.40)
0.79——— 0.83——— 0.83———
Dependency ratiot ’ ’ ’
(0.06) (0.06) (0.06)
1.36 1.74— 1.71—
Relative pricest ’ ’ ’
(0.20) (0.94) (0.95)
Minimum 3.66 6.80 2.49 5.81 2.35 5.90
Maximum 6.31 7.03 5.15 7.20 5.31 7.15
Effect 2.65 2.67 1.37 2.97 1.25
Adjusted R-squared 0.91 0.83 0.92 0.86 0.92 0.86
Number of observations 564 564 564 564 564 564
Signi¬cance levels: — < 0.10; —— < 0.05; ——— < 0.01
Notes: The results for the interactive terms correspond to β in the statistical model. The results for country and time
dummies are not shown.
Coping with Risk

values on the institutional variables.12 The average effect of deindustrial-
ization (shown at the top of the table) is about 4 percent for transfers, and
6 percent for consumption, which accounts for between 60 and 80 percent
of the total time effect (refer to Table 5.2).13
The results for the institutional variables are comfortingly similar to
those for the time dummies. Countries with extensive vocational train-
ing systems and PR always respond to the effects of deindustrialization by
increasing transfer much more than countries with general skills systems
and majoritarian institutions. Countries with PR also tend to increase gov-
ernment consumption more than countries with majoritarian institutions,
whereas there is no effect of vocational training. If we assume that transfers
serve both insurance and redistributive purposes, while government con-
sumption is mainly redistributive, this is precisely the pattern we should
observe. People with speci¬c skills demand insurance while PR promote
alliances behind more spending for both insurance and redistribution. Partisan Effects The ¬nal question to be answered is whether
left and right governments react differently to the employment shocks
produced by deindustrialization. Clearly, we would expect left governments
to show more sensitivity to social needs, and we know from Chapter 4 that
if the electoral system matters for spending, as we just saw, chances are that
partisanship does as well.
The previous analysis included controls for partisanship (higher values
indicate more right-wing governments), but the results were not strong. In
the common shocks version (using time dummies), there is a hint of right
governments spending more than left governments on transfers but not on
consumption. In the deindustrialization version, there is some indication
of left governments spending more on consumption but not on transfers.
Again, if government consumption is more redistributive than transfers,
this is the pattern we should observe. But the more interesting question
is whether left and right governments respond differently to shocks, and
that is indeed what the results in Table 5.5 imply. Left governments are
far more prone to react to deindustrialization by expanding spending than
right governments.

12 Recall that this number is a percentage of the working-age population. As a percentage of
employment, it is more than 10 percent larger.
13 Note that the effect of deindustrialization is not directly comparable to that estimated for
the error correction model. The latter refers to the simulated effect on equilibrium levels
of spending, whereas the former refers to the actual effect in a particular time period.

Forces of Change

Table 5.5. Deindustrialization, Partisanship, and Government
Spending (standard errors in parentheses)

Transfers Consumption
3.90——— 5.17———
Deindustrialization effect
(0.56) (0.45)
’0.92——— ’0.59———
Partisanship — deindustrialization
(0.30) (0.19)
14.98——— 11.91———
(3.77) (3.59)
0.99— 3.44———
GDP per capitat
(0.57) (0.57)
0.81——— ’
Dependency ratiot
’ ’1.15
Relative pricest
Minimum 0.56 3.85
Maximum 3.90 6.29
Range 2.86 2.43
Adjusted R-squared 0.91 0.86
Number of observations 564 564
Signi¬cance levels: — < 0.10; —— < 0.05; ——— < 0.01.
Notes: The results for the interactive terms correspond to β in the sta-
tistical model. The results for country and time dummies are not shown.

At the same time, it must be noted that when shocks are small, the results
suggest that right governments actually spend more than left ones. Why
this is so is an intriguing question. Bear in mind, however, that this is a ¬xed
effects model where the evidence is based entirely on cross-time variation,
and where the temporal relationship between partisanship and spending is
dif¬cult to determine. Chapter 6 will provide a much more detailed account
of partisan responses to deindustrialization.

5.3. The Sources of Deindustrialization
If deindustrialization appears to be such an important force of change, it nat-
urally raises the question of the sources of deindustrialization. Economists
are divided on this question. On one side of the debate, re¬‚ecting not
only a particular economic theory but also a generally popular view (the
“giant sucking sound”), is the idea that the sources of deindustrializa-
tion in the West during recent decades lay in the competitive pressures
Coping with Risk

emanating from Third World producers (Wood 1994; Saeger 1996, 1997).
From this perspective, changes in the North-South trade have been es-
timated to account on average for 50 percent of the reduction in manu-
facturing that occurred between 1970 and 1990 (Saeger 1997, p. 604). In
addition, it can be argued that the removal of restrictions on capital makes
it increasingly easy for businesses to relocate production facilities to coun-
tries with lower wage costs, and that this in turn diminishes the demand
for labor within the industrial sectors of the advanced market economies
(Saeger 1997).
The alternative school, while not denying that trade has played a role
in deindustrialization, sees the principal causes as residing in domestic
sources (Krugman 1996; Rowthorn and Ramaswamy 1997). Among these
are changing preference patterns away from manufactured goods and to-
ward services, high productivity growth in the face of inelastic demand,
as well as the associated changes in investment in new productive capacity
(Rowthorn and Ramaswamy 1999, p. 19). North-South trade accounts for
at most one sixth of the loss in manufacturing employment in these studies.
Furthermore, it may indeed be the case that the welfare state is itself
responsible for the decline in employment in the traditional sectors. As
Bacon and Eltis (1976) have argued, both the costs posed by taxation and
the generosity of the modern welfare state, including the opportunity to
work for at least equivalent if not higher wages in the public sector, have had
a tremendous negative effect on competitiveness and industrial employment
(see also Alesina and Perotti 1997). This is, of course, also a view that is
popular with political parties and governments of a neoliberal bent, but the
discussion in Chapter 4 challenges this idea, and there is little systematic
empirical evidence to support it.
Figure 5.5 provides some descriptive evidence on the question of whether
trade causes deindustrialization. It plots the loss of employment in the tradi-
tional sectors from 1962 through 1991 against the average trade openness
for the same period. There is little hint of any relationship. Indeed the
correlation between the two series is about 0.17.
Alternatively, if one were to adopt the hypothesis that deindustrialization
has more to do with internal processes (processes of productivity gain and
shifting tastes), then one would expect that a process of convergence has
been underway. Thus, early industrializers, which had pretty much gone
through this transformation by the beginning of this period, would have
suffered the least loss of employment in the traditional sectors, while late
industrializers would have experienced more rapid decline.
Forces of Change

Drop in size of traditional sectors, 1962“91

20 France Denmark
Norway Belgium
15 Sweden Austria
Australia Netherlands



0 50 100 150
Trade openness, 1962

Figure 5.5 Trade openness and losses in traditional sectors.

As Figure 5.6 demonstrates, there seems to be a fair amount of sup-
port for this position. The correlation between employment intensity in
the traditional sectors in the year 1962 and the loss of employment in these
sectors over the three succeeding decades is about .85. Thus, the United
States, which had the smallest traditional sectors (about 24 percent), expe-
rienced the smallest loss (less than 5 percent), while Finland, lagging well
behind the United States and having nearly 50 percent of its working-age
population engaged in the traditional sectors, experienced the largest loss
in the sample of ¬fteen countries, well over 20 percent.
But descriptive, and indirect, evidence of this nature can be misleading,
and I have therefore estimated a pooled cross-sectional time-series model
that uses the change in the log of the number of people employed in man-
ufacturing and agriculture as a share of the working-age population as the
dependent variable (see Table 5.6).14 This is a standard setup in the exist-
ing literature except that I have included agricultural employment on the
left-hand side to make the results speak directly to the deindustrialization
variable. However, the results are very similar if one focuses exclusively on

14 As in the previous analysis, problems of heteroskedasticity led us to employ Beck and Katz™s
(1995) method for deriving panel corrected standard errors.

Coping with Risk

Drop in size of traditional sectors, 1962“91

20 France Italy Denmark
UK Germany Austria
Norway Sweden



20 25 30 35 40 45 50
Initial size of traditional sectors, 1962
(Percent of working-age population)

Figure 5.6 Initial size and losses in traditional sectors.

manufacturing employment. The analysis includes fourteen OECD coun-
tries for which there are complete data in the period from 1964 through
For presentational ease, Table 5.6 divides the independent variables into
a group of “domestic” variables and a group of “international” variables.
Following the existing economic literature, the domestic-structural vari-
ables include (i) a measure of productivity growth, (ii) the log of income per
capita and the square of this variable to capture changing consumption pref-
erences, (iii) the growth in per capita income as a measure of demand effects,
(iv) gross capital formation as a share of GDP, and (v) the two spending vari-
ables. For the exogenous variables the regression includes (vi) the balance
of trade with OECD, the Organization of Petroleum Exporting Countries
(OPEC), and less-developed countries (LDCs), and (vii) the Quinn-Inclan
capital market openness variable.

15 The countries include Austria, Belgium, Canada, Denmark, Finland, France, Germany,
Italy, Japan, Netherlands, Norway, Sweden, the United Kingdom, and the United States.
Missing data problems precluded adding Switzerland. The time frame is the maximum
possible given the availability of data.

Forces of Change

Table 5.6. Regression Results for Industrialization (t-scores in parentheses)

Endogenous Variables Exogenous Variables
[Lagged level] Capital openness 0.001
(’5.27) (0.30)
Productivity growth Capital openness 0.001
(9.09) (0.50)
0.523—— 0.002———
Income OECD trade balance
(2.18) (3.47)
’0.30—— 0.004———
Income squared OECD trade balance
(2.24) (4.43)
0.585——— ’0.004—
Growth in income OPEC trade balance
(8.50) (’1.96)
0.032——— ’0.003
Capital formation OPEC trade balance
(5.53) (’1.35)
Government transfers LDC trade balance
(’0.99) (’2.12)
Government 0.001 LDC trade balance
(0.30) (’1.19)
Consumption (0.30) (’1.19)
Increase in explained variance 35% 5%
Adjusted R-squared 0.52
Number of observations 378 378
Signi¬cance levels: — < 0.10; —— < 0.05; ——— < 0.01.
Notes: The increase in explained variance is the change in R-squared when the set of en-
dogenous and exogenous variables are added to a model where these variables are excluded.
The results for country dummies are not shown.

The productivity measure is meant to capture the tendency for ¬rms to
shed workers as productivity increases. Note that there is some theoretical
ambiguity with respect to the impact of this variable. Even though faster
productivity growth makes goods relatively cheaper, and therefore boosts
demand, less labor is required to produce the same amount of output. The
net effect on employment depends on the price and income elasticity of
demand, as well as on real wage changes. Research, however, has shown that
the labor-saving effect tends to dominate the demand effect in the period of
interest (Appelbaum and Schettkat 1994, 1995). For the income terms, the
expectation is that the parameter on the ¬rst term will be positive while that
on the second term will be negative, signifying that as income passes beyond
a certain level, the relative demand for goods in both the agricultural and
manufacturing sectors will begin to decline. The effects of capital formation
Coping with Risk

and growth in income are expected to be positive because both will boost
production and demand for labor.16
The results show that deindustrialization appears to be driven mostly by
domestic factors other than the welfare state. Technological progress, de-
mand conditions, and shifting consumption patterns cause employment in
industry and agriculture to decline. There is no evidence that government
spending has “crowded out” employment in the traditional sectors; every
indication is that the causal arrow goes in the opposite direction. Nor does
trade appear to be a particularly important source of deindustrialization.
A negative trade balance with other industrialized countries (and the ¬rst
difference in that trade balance) does hurt industrial employment, but the
effect is substantively small and cannot have been a major cause of deindus-
trialization across the OECD area for the simple reason that intra-OECD
trade is relatively balanced over time.
The crucial question with respect to trade is whether growing trade with
less-developed countries has priced out a substantial number of workers in
agriculture and industry in the advanced countries. There is no evidence to
that effect. The coef¬cients on the lagged levels of the trade balances with
OPEC countries and with Third World countries are both negative and
statistically signi¬cant, while both of the coef¬cients on the ¬rst differences
are statistically insigni¬cant. Note that these results, which suggest that
positive trade balances with the OPEC and Third World countries lower
employment, while negative balances promote employment, are not the
consequence of multicollinearity. Nor do their effects change in substantive
terms when one uses alternative speci¬cations of the model. A large number
of regressions using a variety of combinations of trade balances and import
penetration measures were run, and the results are all contrary to the “trade
leads to deindustrialization” hypothesis. In fact, the results presented in
Table 5.6 are the strongest possible in support of the popular perception. The
same is the case for the capital market openness variable, which consistently
fails to produce effects that are statistically distinguishable from zero.
Though somewhat surprising given popular views, the results essentially
replicate those in an IMF study by Rowthorn and Ramaswamy (1999), even
though data and model speci¬cation vary somewhat. As in the Rowthorn-
Ramaswamy study, deindustrialization is driven primarily by economic

16 Investment is measured as a percentage share of GDP. It is taken from Robert Summers and
Alan Heston, Penn World Tables, Version 5.5, data ¬le (Cambridge, MA: National Bureau
of Economic Research, 1993).

Forces of Change

processes that are unrelated to either openness or spending. Productiv-
ity growth in the traditional sectors leads to a loss in employment, whereas
rising demand through growing investment or incomes has a positive effect.
Consistent with Engel™s law, the results also indicate that demand for agri-
cultural and manufacturing ¬rst rises with income and then falls at higher
levels, thereby eventually diminishing the level of traditional employment.
Hence, the argument and results for spending appear quite robust to both
the charge that deindustrialization is merely a mediating variable and the
charge that its association with spending is a result of reversed causality.
One caveat is that technological change and productivity growth are
affected by international competition. Undoubtedly, there is some truth


. 7
( 11)