The Long and Short (of) Quality Ladders∗

Amit Khandelwal†

Columbia Business School

First Draft: November 2006

This Version: May 2008

Abstract

This paper offers evidence that a market’s scope for quality differentiation, or its “quality ladder,”

can explain the heterogenous impact of low-wage competition on U.S. industry employment and output.

I construct a market’s quality ladders by first estimating countries’ export quality to the U.S. using

a procedure that exploits both price and quantity information, which contrasts to earlier work that

relies only on price data. Markets differ in their estimated quality ladder lengths implying that the

U.S. exhibits heterogeneity in its exposure to low-wage competition. Empirical estimates confirm that

the impact of low-wage import penetration on U.S. manufacturing output and employment is weaker in

markets characterized by longer quality ladders.

Keywords: Quality Ladders; Low Wage Competition; Quality Specialization; Product Differentiation

JEL Classification: F1, F15, F16

1 Introduction

The fear of globalization’s impact on domestic employment is rooted in the vulnerability or, to use Edward

Leamer’s terminology, the contestability of jobs (Leamer, 2006). As Leamer puts it, the contestable jobs are

those whose “wages in Los Angeles are set in Shanghai.”1 Recent attention in the media and political arena∗I am especially grateful to my dissertation committee, Irene Brambilla, Penny Goldberg and Peter Schott, for guidance and

support. I have benefited from conversations with Steve Berry, Ray Fisman, Juan Carlos Hallak, David Hummels, Kala Krishna,

Chris Ksoll, Frank Limbrock, Nina Pavcnik, Siddharth Sharma, Gustavo Soares, Robert Staiger, Catherine Thomas, Daniel

Trefler, Chris Udry, David Weinstein, Jeffrey Weinstein, and various seminar participants. Special thanks also to Amalavoyal

Chari. All errors are my own.†Uris Hall 606, 3022 Broadway, New York, NY 10027, email: ak2796@columbia.edu, website:

http://www0.gsb.columbia.edu/faculty/akhandelwal/.1Leamer (2006), page 5.

1

has elevated these fears by arguing that increased economic integration would result in today’s American

jobs being “exported abroad” tomorrow.

A recent study by Bernard, Jensen, and Schott (2006) provides evidence that the probability of

U.S. plant survival and employment growth are negatively associated with an industry’s exposure to import

penetration, particularly from low-wage countries.2 But while low-wage competition negatively affects output

and employment growth, the impact appears heterogenous across industries. For instance, between 1980 and

the mid-1990s, electronics (SIC 36) experienced greater low-wage import penetration than fabricated metals

(SIC 34) but experienced a smaller employment decline.3

This paper provides an explanation for the heterogenous response that goes beyond the textbook

factor proportions framework. In a standard HO model, labor-intensive markets would be relatively more

affected by low-wage competition than capital-intensive markets. In my framework, the impact of low-

wage competition on employment varies with a market’s degree of quality specialization.4 This argument

is related to Schott (2004) who finds evidence against specialization across products (e.g., capital- versus

labor-intensive products) in favor of specialization within products along the quality dimension (e.g., low-

and high-quality varieties within products). In particular, developed countries are found to export more

expensive varieties within detailed products relative to developing countries.

If countries inhabit different cones of (quality) diversification, developed countries are theoretically

insulated from low-wage countries. However, I argue that if markets vary in their scope for quality differ-

entiation, developed countries will experience heterogeneity in their exposure to developing countries. For

example, developed countries may face relatively more competition in cotton shirts than say, computer mon-

itors, if quality differentiation for shirts is less feasible. Variation in markets’ “quality ladders”, defined as

the range of qualities within a market, will arise if the benefit and/or cost of increasing quality varies across

markets. And while the quality ladders certainly evolve over time, at least in the short- to medium-run, a

market’s quality ladder therefore may be characterized as long or short.

If the range of equilibrium qualities within a market is large, developed countries can insulate

themselves from the South by using comparative advantage factors (e.g., skill, capital and/or technology)

to specialize atop the quality ladder. In markets characterized by short quality ladders, developed countries2Other studies studying the negative relationships between trade and employment include Sachs and Shatz (1994), Free-

man and Katz (1991) and Revenga (1992). Bernard, Jensen, and Schott (2006) explicitly connect the relationship between

employment and trade with low-wage countries, defined as nations with less than 5 percent of U.S. per capita GDP. I use their

definition of low-wage countries in this paper (see Table 1).3One potential explanation is differences in capital intensity, but in 1980, electronics was less capital intensive than fabricated

metals. Indeed, this paper offers evidence that capital intensity only partly explains heterogeneity in U.S. employment and

output responses due to import competition.4The argument is related to studies that reject the textbook version of the Heckscher-Ohlin model which predicts factor price

equalization (FPE) when countries trade identical goods using identical technologies. See Leamer (1987) and Schott (2003).

2

will be directly exposed to Southern competition. A market’s scope for vertical differentiation is therefore

important for understanding market contestability: the impact of production in countries like China and

India on U.S. employment and output will depend critically on the length of a market’s quality ladder.

I assess this quality ladder hypothesis in two stages. First, I obtain empirical measures of import

quality. This has posed a challenge to the literature since most studies simply invoke a convenient vertical-

market assumption which enables observed unit values to perfectly proxy for unobserved quality.5 However,

an import may be expensive not because it has high quality but because factor prices in the originating

country are high. A consumer may purchase this expensive, but low quality, import nonetheless if she has

an idiosyncratic preference for the product’s horizontal attribute.

Rather than assuming that prices perfectly proxy for quality, I estimate quality measures derived

from a structural model of consumer demand that embeds preferences for both horizontal and vertical

attributes. I estimate nested logit demand curves, as in Irwin and Pavcnik (2004), for each of approximately

1,000 manufacturing industries. Quality is the vertical component of the estimated model and has a structural

definition as the mean valuation that U.S. consumers attach to an imported product. The intuition behind

this approach is similar to Hallak and Schott (2007): conditional on price, the product with higher market

share is assigned higher quality–and a key advantage of the nested logit procedure used in this paper is that

it recovers quality at the finest level of product aggregation available in the data.

The inferred qualities reveal that exports to the U.S. from developed countries sit atop the quality

ladders, defined as the range of estimated qualities within a ten-digit HS product. The quality ladders,

however, vary in their length implying that developed countries’ export quality is relatively higher than

developing countries in some markets. In other markets, such as apparel and footwear, the quality ladders

are compressed despite variation in prices suggesting that expensive imports can coexist with cheaper rivals

due to horizontal product differentiation. In these short-ladder markets, a fraction of consumers value the

product’s horizontal attribute but the average U.S. consumer attaches a low valuation to the expensive

import. This underscores the potential danger in invoking the “price = quality” assumption, particularly in

products characterized by short quality ladders.

Having recovered quality from U.S. imports and constructed quality ladders, I then aggregate the

product-level quality ladders to match U.S. industry data in order to test the quality ladder hypothesis.

Consistent with Bernard et al. (2006), I find that industry employment is negatively associated with exposure

to imports, particularly from low-wage countries. My hypothesis, formally outlined in Section 2, predicts

a differential impact across product markets according to ladder length. The empirical results confirm

that import penetration has a weaker impact on employment in industries with long quality ladders: a ten

percentage point increase in low-wage penetration is associated with a 6.5 percent employment decline in an5E.g., see Brooks (2003), Schott (2004), and Hallak (2006).

3

industry characterized by an average quality ladder length. A similar increase in competition in a long-ladder

industry (one standard deviation above the mean) results in only a 1.6 percent employment decline. The

regressions are robust to a range of controls and specifications and also to instrumenting for endogenous

import competition.

The results of this paper contribute to the growing literature that recognizes the importance of

quality specialization in international trade. One line of research has focused on introducing quality into

heterogenous-firm models in order to better match theoretical predictions with observed trade patterns (e.g.,

see Kugler and Verhoogen (2008)). A separate line of research has focused on quality specialization from the

perspective of developing countries. In this work, quality upgrading has important implications for export

success and, ultimately, economic development (e.g., Schott (2008), Rodrik (2006), Hidalgo et al. (2007) and

Verhoogen (2008)). Quality specialization also explains why inequality may be rising in developing countries,

in contrast to standard Stolper-Samuelson predictions, as they liberalize their economies (e.g., Goldberg and

Pavcnik (2007) and Verhoogen (2008)). One contribution of this paper is to offer a simple way to estimate

quality that is robust to horizontal differentiation at a finely disaggregated level. These quality measures

can easily be incorporated into these two lines of research.

This paper also provides insight into the role of quality specialization in determining the impact of

foreign competition on domestic industries. Previous work, such as Sachs and Shatz (1994) and Bernard

et al. (2006), has documented heterogenous impacts based on traditional Heckscher-Ohlin channels via

differences in industry skill and/or capital intensity. Here, I add quality ladders as an additional explanation.

Importantly, the impact of import competition on short and long-ladder industries is similar in magnitude

to the differential impact on low and high capital-intensive industries. In other words, even after controlling

for the differential impact according to capital and skill intensities, the quality ladder remains an important

determinant of an industry’s vulnerability to low-wage competition. As a result, a quality upgrading strategy

may be only effective for some markets as a response to increased competition from countries like China and

India. Quality ladders therefore help identify which markets are likely to be relatively more susceptible to

increasing foreign and low-wage competition.

The remainder of the paper is organized as follows: Section 2 offers a simple model to illustrate that

exposure to low-wage competition is greater in markets with short quality ladders. The empirical method

used to identify quality in the data is discussed in Section 3. The data and quality estimation results are

presented in Section 4. Section 5 applies the quality ladders to U.S. industry data to test the implications

of quality specialization for employment and Section 6 concludes.

4

2 A Simple Model

This model formalizes the hypothesis that U.S. firms will be less vulnerable to foreign competition in in-

dustries characterized by longer quality ladders. The model delivers two comparative static results that are

verified in the empirical analysis below. Consider firms in two regions, North (N) and South (S), where the

Southern firms freely export to the North (no trade costs). This partial equilibrium model analyzes prices,

qualities and market shares in the North’s (domestic) market. Wages are taken to be exogenous and are

assumed higher in the North than the South: wN > wS .

I abstract away from firm heterogeneity (e.g., Melitz (2003)) by assuming that within each region

there are J homogenous firms. Firms compete by manufacturing horizontally and vertically distinct varieties.

Following Krugman (1980) and Melitz (2003), horizontal differentiation is costless so in equilibrium, all

firms produce horizontally distinct varieties. But as in Flam and Helpman (1987), vertical (e.g., quality)

differentiation depends on a Ricardian-type comparative advantage. All firms face the same marginal cost

curve within the region. Marginal cost of production is constant but increases with quality (λi): wi + λ2i

2Zi

for i ∈ {N, S}. The parameter Zi denotes region i’s access to technology and I assume that Northern firms

have access to better technology than the South: ZN > ZS . Firms pay a fixed cost of production Fi.

The consumers who live in the North have discrete choice preferences. Consumer n observes the

domestic and Southern varieties and chooses the variety j that provides her with the highest indirect utility

Vnj = θλj − pj + εnj . (1)

Quality is defined as an attribute whose valuation is agreed upon by all consumers: holding prices fixed, all

consumers would prefer higher quality objects. There are several interpretations of this vertical component.

It can measure observable characteristics like the clarity or sharpness of a television screen or it can reflect the

perceived quality that results from advertising. In either case, quality represents any attribute that enhances

consumers’ willingness-to-pay for a variety. A more straightforward interpretation is that λ represents a

shift parameter in the variety’s demand schedule: holding price pj fixed, demand shifts out when the quality

improves (Sutton, 1991). The empirical identification of quality relies on this latter intuition. Note that

since firms within a region face the same cost function, each firm j in region i chooses the same quality:

λj = λi, ∀j ∈ i. In the remainder of the analysis, I drop the subscript j unless necessary.

The parameter θ reflects the consumers’ valuation for quality and plays an important role in this

model. The comparative statics below analyze how the models’ endogenous variables adjust if θ varies

across markets. One could also derive comparative statics with respect to differences in the marginal cost of

quality across markets (which is fixed in this model). However, the empirical analysis identifies differences

in consumer valuations rather than differences in costs, so analyzing differences in markets in terms of θ

represents a closer link between the model and the empirical analysis. It will be convenient to define the

5

mean valuation for the representative variety in region i, inclusive of price, as δi ≡ θλi − pi.

Horizontal product differentiation is introduced in (1) through the consumer-variety-specific term.

This term explains why some consumers purchase varieties with high quality-adjusted prices. For example, a

consumer may purchase an expensive shoe that is uncomfortable if she has a particular affinity for the variety

in terms of its design or style.6 If εnj is i.i.d. type-I extreme value, the market share for the representative

firm in region i is given by the familiar logit formula

mi =eδi

∑j∈J eδN +

∑j∈J eδS

, i ∈ {N,S}

=eδi

J (eδN + eδS )(2)

Firms maximize profits in the Northern market by choosing price and quality by solving the following

problem

maxpi,λi

[pi − wi − λ2

i

2Zi

]eδi

J (eδN + eδS )− Fi (3)

With monopolistic competition, each firm is unable to influence the market which implies that the denom-

inator in (2) is fixed. As shown in Anderson et al. (1992), the pricing solution to (3) under monopolistic

competition is

p∗i = 1 + wi +λ∗2i

2Zi(4)

This expression is the standard markup over marginal cost where the markup in this example is 1.7 The

optimal quality choice by the representative firm in i is

λ∗i = θZi (5)

The expression in (5) indicates that quality choice is increasing in consumers’ valuation of quality and firms’

access to technology. Since the North has better technology, Northern firms choose higher quality than

Southern firms.8

From (2), the aggregate market share of firms in region i is Mi ≡∑

j∈i mi = Jeδ∗i

J(

eδ∗N +eδ∗

S

) where the

mean valuation for the representative firm at the optimal price and quantity is

δ∗i =θ2Zi

2− 1− wi (6)

6Comfort is a quality attribute since, ceteris paribus, all consumers prefer more comfortable shoes. A shirt’s design is a

horizontal attribute since some consumers, for instance, prefer stripes over solids.

7Modifying the indirect utility function to Vnj = θλj − αpj + εnj yields a price equal to 1α

+ wi +λ∗2i2Zi

. If consumers have

perfectly elastic demand (α = ∞) prices converge to marginal cost.8I should note that since p∗N > p∗S , prices are sufficient statistics in this model This occurs since horizontal attributes are

assumed to be costless. The model could easily incorporate an additional cost for horizontal attributes which could break the

one-to-one mapping between prices and quality.

6

It is reasonable to assume parameter values such that aggregate market share in the North exceeds the

South:

eδ∗N > eδ∗S =⇒ θ2

2(ZN − ZS) > (wN − wS) (7)

This assumption is based on substantial theoretical and empirical work arguing that higher quality, or more

productive, firms have relatively higher output.9 The intuition behind condition (7) is that the return

to the North’s technological prowess over the South is sufficient to overcome the North’s disadvantage in

manufacturing costs.10

Following Grossman and Helpman (1991), I define the quality ladder as the difference between the

highest and lowest equilibrium quality. From (5), the ladder for a market with consumer valuation θ is

Ladder(θ) ≡ λ∗N − λ∗S = θ (ZN − ZS) (8)

The ladder represents a market’s scope for quality differentiation. Now, consider two markets that are

otherwise equal but differ only in consumers’ valuation for quality, θ′ > θ. Equation (8) trivially shows that

as the marginal benefit for quality increases, the quality ladder increases, or “lengthens”. Moreover, the

gains are disproportionately distributed to the manufacturers of higher quality (∂δ∗N∂θ >

∂δ∗S∂θ ).

This model abstracts away from the endogenous “lengthening” of the ladder that may occur in a

long-run equilibrium with technological progress or shifts in consumer preferences.11 Instead, this model

generates a quality ladder that is fixed in the short- and medium-run. In the empirical analysis below,

I mitigate endogeneity concerns by assigning a market’s quality ladder its initial length. However, there

is a high correlation between a market’s initial ladder length and its length at the end of the sample. In

other words, on average, markets with initially “short” ladders are not “long” by the end of the sample. This

implies that the quality ladder is an intrinsic feature that measures a market’s scope for quality differentiation

and endogeneity issues may be less of a concern.

Having established market equilibrium, I now compute two comparative static predictions. The first

establishes that the North’s aggregate market share is influenced by the nature of its Southern competition.

As manufacturing wages in the South (exogenously) decline, the North becomes relatively less competitive

and loses market share. The change in North’s market share with respect to change in the type of foreign

competition is given by∂MN

∂wS= −MNMS

∂δ∗S∂wS

> 0, (9)

since from (6), ∂δ∗S∂wS

= − θ2ZS

2w2S

< 0. Firms become more competitive and increase market shares as manufac-

turing costs fall. This comparative static is supported by existing empirical evidence. Bernard et al. (2006)9For instance, see Melitz (2003), Bernard et al. (2007) or Verhoogen (2008).

10For example, if θ =√

2, the condition states that while the North has higher wages, its productivity-adjusted wages are less

than the South.11See, for instance, Aghion and Howitt (1992), Sutton (1998) or Klette and Kortum (2004).

7

show that output and employment for U.S. plants are negatively associated with import competition, but

the impact is much larger when import competition originates from countries with less than 5% of U.S. per

capita GDP. Conditional on quality, competition from China should have a larger affect on U.S. firms than

competition from Japan.

Importantly, this model adds quality differentiation to the analysis to show that the intensity of

competition within a market depends on the quality ladder length. In particular, while (9) indicates that

the North’s market share falls as Southern wages decline, it suffers a smaller market share loss in markets

characterized by longer quality ladders. There is a monotonic relationship between the ladder length and θ

so the cross-partial derivative of (9) with respect to θ is:

∂2MN

∂wS∂θ= −

[∂δ∗S∂wS

(MS

∂MN

∂θ+ MN

∂MS

∂θ

)](10)

= −[MNMS

∂δ∗S∂wS

(MS −MN )(

∂δN

∂θ− ∂δS

∂θ

)]

= −θMNMS (ZN − ZS) (MN −MS) < 0,

because of condition (7) and the assumption that ZN > ZS . This derivative states that in long-ladder

markets (high θ), the sensitivity of Northern market shares to Southern competitiveness is reduced. So, a

decrease in the South’s wage results in a smaller decline of the North’s market share in long ladders.

The model shows that trading with the South can generate a differential impact on two markets that

are otherwise identical but vary according to consumers’ valuation for quality. This result is related to more

general trade models that predict a breakdown of FPE when countries are fully specialized in production.

In contrast to a single-cone equilibrium, where endowments are such that all countries produce all goods,

the conditions required for factor price equalization are not met in multi-cone equilibrium because countries

specialize in varieties tailored to their endowments.12 Schott (2004) has extended this analysis to within

product specialization where endowment differences cause countries to specialize in different segments of

a product’s quality ladder. The model here sharpens this analysis by arguing that the scope for quality

specialization varies across markets. For example, multiple cones may not arise in the glass vase market but

could arise in the transmission receivers market with developed countries specializing in digital receivers and

developing countries manufacturing analog receivers. This would imply that U.S. vase producers compete

directly with their Chinese counterparts while U.S. electronics firms reduce exposure because they inhabit a

higher rung of the television quality ladder.12For evidence in favor of the hypothesis that countries inhabit multiple cones of diversification, see Leamer (1987), Davis

and Weinstein (2001) and Schott (2003).

8

3 Empirical Implementation

In order to test the two comparative static predictions in (9) and (10), I structurally implement the demand

system to obtain empirical measures of markets’ quality ladders. I obtain ladders by first estimating the

quality of U.S. imports and then constructing the ladder for each ten-digit HS product. The methodology

used to estimate quality is a nested version of the logit demand specification in (1). This demand system

is more flexible than the logit model above. The nested logit partially relaxes the IIA property by allowing

preferences for alternatives within a nest, such as the type of clothing material, to be more correlated with

each other. Accounting for IIA is important because I infer quality from both price and market share

data. To understand why, suppose a consumer chooses between a Japanese wool shirt and an Italian cotton

shirt. Under a standard logit or CES framework, the market shares and inferred consumer valuation of

both imports fall by an equal percentage if a Chinese cotton shirt enters the market. However, the Chinese

shirt’s “location” should be closer to the Italian shirt because of the similarity in material. The market share

of the Italian shirt should adjust by more than the Japanese wool shirt. The nested logit allows for more

appropriate substitution patterns by placing varieties into appropriate nests and is conceptually equivalent

to a nested CES.

I estimate separate demand curves at the five-digit Standard Industrial Trade Classification (SITC,

Revision 2) level. The SITCs are aggregates of ten-digit Harmonized System (HS) products. For example,

within the men’s knit-shirts industry (SITC 84632) there are different types of shirts by HS products (e.g.,

cotton, wool, and silk). The HS codes therefore provide a natural delineation for the nests because the

product descriptions classify imports along similar characteristics. A country c’s export within a HS product

k is referred to as a variety.

The following derivation of consumer preferences is for a single SITC industry. Consider consumer

n’s preferences over country c’s export of HS product k (a variety) within an SITC industry at time t. The

consumer purchases the one variety that provides her with the highest indirect utility, given by

Vnckt = λ1ck + λ2t + ∆λ3ckt − αpckt +K∑

k=1

µnktdck + (1− σ)εnckt. (11)

The λ terms represent the variety’s valuation that is common across consumers (these terms are not sub-

scripted by n). This is the empirical analog to quality in the theoretical model and is decomposed into three

components. The first term, λ1ck, is the time-invariant valuation that the consumer attaches to variety ck.

The second term, λ2t, controls for secular time trends common across all varieties. The ∆λ3ckt term is a

variety-time deviation from the fixed effect that is observed by the consumer but not the econometrician.

This term is potentially correlated with the variety’s c.i.f. unit value, pckt (inclusive of transportation and

tariff costs).

9

The horizontal component of the model is captured by the logit error εnckt and∑K

k=1 µnktdck. This

latter term interacts the valuation that consumer n places on product k, µnkt, with a dummy variable dck

that takes a value of 1 if country c’s export lies in product k. This term enables correlations among consumer

n’s preferences for all varieties within product k.13

An outside variety completes the demand system. The consumer chooses the outside option if it

provides more utility than all of the inside varieties. The outside option captures the utility for purchasing

a domestic U.S. variety or not purchasing any variety. The utility of the outside option is given by

un0t = λ10 + λ2t + λ30t − αp0t + µn0t + (1− σ)εn0t, (12)

and is normalized to zero (see Berry (1994) for details).

Since the outside variety market share is unobserved, I proxy it by using import penetration measures

at the four-digit Standard Industrial Classification (SIC, Revision 1987), which are taken from Bernard et al.

(2006) and mapped to the SITC industries using a concordance provided by Feenstra et al. (2002). The

outside variety market share is defined as one minus the import penetration ratio. It is important to note that

the assumption on the outside good market share will not affect the results below because the specification

includes year fixed effects which are common across all varieties within the SITC industry. The total market

size for the industry can then be obtained from MKTt =∑

c∈Jkt,c6=0 qckt/(1 − s0t), where qckt denotes the

import quantity of variety ck. The imported variety market shares are computed as sckt = qckt/MKTt. In

other words, while the outside good market share affects the absolute growth rate of quality, the relative

quality growth rate is unaffected.

The consumer chooses variety j if Vnckt > Vnc′k′t, ∀ c′ 6= c, ∀k. Under the distributional assumptions

for the random component of consumer utility, Berry (1994) has shown that the demand curve implied by

the preferences in (11) is

ln(sckt)− ln(s0t) = λ1ck + λ2t − αpckt + σ ln(s̄ct/k) + ∆λ3ckt, (13)

where s̄ct/k is variety ck’s share within product k at time t.14 The term s0t is the outside good market share.

The expression in (13) states that the relative market share of the variety will equal its mean

valuation plus its significance within the nest it occupies, less its price. As discussed above, the nest share

term (s̄ct/k) controls for “crowding” within an HS by allowing for stronger correlations in preferences for

13Cardell (1997) has shown that the distribution of∑K

k=1 µnktdck is the unique distribution such that if ε is distributed

extreme value, then the sum is also distributed type-I extreme value. The degree of within nest correlation is controlled by

σ ∈ (0, 1] and is assumed to be identical across all products. As σ approaches one, the correlation in consumer tastes for

varieties within a nest approaches one and as σ tends to zero, the nested logit converges to the standard logit model.14This term is a consequence of the nested logit structure. If one adopts a standard logit utility function, this term disappears

from equation (13).

10

varieties within nests.15 λ1ck + λ2t are usually approximated using the variety’s characteristics but the

import data do not record characteristics. I therefore exploit the panel dimension of the data by estimating

variety (country-product) and year fixed effects. The third component of quality, ∆λ3ckt, is not observed

and plays the role of the estimation error. Both this term and the nest share are potentially correlated with

the variety’s price, so instrumental variables are required to identify the parameters.

3.1 Identification and Hidden Varieties

Identifying the price coefficient in (13) typically relies on rival variety characteristics as instruments (Berry

et al., 1995). Yet even if variety characteristics were available in the import data, the assumption of exogenous

characteristics may be problematic if firms simultaneously choose prices and characteristics (as in the model).

Fortunately, the import data provide variety-specific transportation costs that can serve as instruments

for the c.i.f. price. Transport costs are obviously correlated with c.i.f. prices, but may potentially be

correlated with prices if firms ship higher quality goods in order to lower per unit transport costs (the

Alchian-Allen Conjecture). This potentially raises concerns that trade costs may be correlated with variety

quality. However, the exclusion restriction remains valid as long as transport costs affect average quality λ1ck

and not the time-specific deviation, ∆λ3ckt. Moreover, I also instrument using exchange rates and distance

to the U.S. interacted with oil prices. Including these instruments further mitigates concerns that transport

costs influence quality choice by partialling out the effect of exchange rates and oil shocks on prices in the

first stage.

The nest share term on the right-hand-side of (13) is also endogenous, so I use the number of varieties

within the nest and the number of varieties exported by the country to identify σ. The validity of these

instruments relies on weaker assumptions than is often made in the discrete choice literature. This literature

assumes a fixed variety space (both variety characteristics and number of varieties) and does not model entry

or exit (Berry et al. (1995)). As a result, the number of varieties satisfies the exclusion restriction when

entry and exit and varieties occur prior to the revelation of ∆λ3ckt.

A second issue that arises in estimating (13) is that the market shares are likely to be an aggregation

of even more finely classified imports. As noted in Feenstra (1994), a country’s large market share may

simply reflect the fact that it exports more unobserved or hidden varieties within a product. To illustrate

this potential problem, suppose that China and Italy export identical varieties at identical prices and split

the market equally at the (unobserved) twelve-digit level, but that China exports more twelve-digit varieties

(such as more colors). Aggregation to the observed ten-digit level would assign a larger market share at

identical prices to China. From (13), China’s estimated quality would be biased upward bias simply due

15Under log-linearity in income and prices, Anderson et al. (1987) have shown that the demand curves generated by the logit

and CES are identical. Thus, the specification derived here is analogous to a nested CES.

11

to the horizontal varieties. In certain contexts it might be appropriate to include unobserved varieties in

the quality measure. But the theory here emphasizes the importance of vertical, rather than horizontal,

specialization, and so it becomes necessary to control for within-product horizontal differentiation in this

context. Allowing the mean of the logit error distribution to vary across countries is one method to control

for the unobserved varieties, but country fixed effects are not identified in (13) because of the variety fixed

effects. I therefore follow theoretical predictions in Krugman (1980), among others, and use a country’s

population as a proxy for the number of hidden varieties. Allowing the (log of) population to shift the logit

error mean implies that it becomes an additional covariate in (13).16

The demand curve that corrects for the hidden varieties is given by

ln(sckt)− ln(s0t) = λ1ck + λ2t − αpckt + σ ln(s̄ct/k) + γ ln popct + ∆λ3ckt, (14)

where popct is the population of country c. I estimate equation (14) separately for each of the approximately

1,000 SITC industries. The quality of variety j (recall that j is a country-product pair) at time t is defined

as

λckt ≡ λ̂1ck + λ̂2t + ∆̂λ3ckt. (15)

Previous work in the international trade literature has assumed that observable unit values can perfectly

proxy for unobserved quality (one exception is Hallak and Schott (2007)). As discussed earlier, this as-

sumption is problematic if consumers have preferences over horizontal attributes. The demand structure

used in this paper explicitly allows varieties to possess both types of product differentiation. Equation (14)

shows that quality is the residual of the demand system and captures the consumers’ willingness-to-pay.

The intuition behind (15) is that an increase in a product’s quality allows its price to rise without losing

market share. It is possible that factors could affect market shares but it is important to note that this set

is made much smaller by conditioning on prices. For example, a variety may have a large market share if

the exporting country is geographically close to U.S. However, the price includes transportation costs and

therefore the quality estimate is not capturing purely gravity “forces” such as distance. A similar argument

can be made regarding free trade agreements: even though Mexican and Canadian import shares are high

because of NAFTA, this effect will operate through prices, which are inclusive of tariffs. Likewise, a low-wage

country may have high market shares, but by conditioning on its low export price, the quality measures will

not just reflect market shares.

4 Data and Quality Estimation Results

I estimate the regression in (14) on U.S. product-level import data from 1989-2001 (Feenstra et al., 2002).

The sample is restricted to the manufacturing industries (SITC 5-8) and excludes the homogenous goods16The results are not sensitive to using total GDP as an alternative proxy. See below.

12

defined by Rauch (1999) since these products, by definition, exhibit no quality differentiation. In addition

to import quantities and values, the data record tariffs and transportation costs. A variety’s unit value is

defined as the sum of the value, total duties and transportation costs divided by the import quantity and

deflated to real values using the CPI.

Since the import data are extremely noisy (General Accounting Office, 1995), I trim the data along

two dimensions. The first trim excludes all varieties that report a quantity of one unit or a total value of less

than $10,000. The second trim removes varieties with extreme unit values that fall below the 5th percentile

or above the 95th percentile within an SITC industry.17 Table 2 reports basic summary statistics by two-

digit SIC sectors. The average apparel and leather imports originate from countries with lowest average

per capita income while imports into transportation, industrial machinery and chemicals are dominated by

relatively richer countries. Notice also that there is little variation in the number of countries present at the

two-digit level which indicates the lack of across-sector specialization among U.S. trading partners noted by

Schott (2004).

The estimating equation in (14) is run separately for 1,059 SITC industries and summary statistics

of the regression results are reported in Table 3. The standard errors are clustered by exporting country to

allow for correlations between varieties and over time. The bottom panel reports that 70 percent of the 1,059

regressions, which amounts to 80 percent of the 1.2 million observations, have a negative and statistically

significant price coefficient. Rows one and two in the top panel indicate that the average IV price coefficient

is about 72 percent lower than the OLS price coefficient. Row 3 reports the price elasticity obtained based

on α̂. The average own price elasticity is -1.54, which is low, but this is not surprising given that (14)

estimates the parameters using within variety variation. Rows 4-6 indicate that the average and median

regression pass the overidentifying restrictions test and have low first-stage F-statistic p-values. 56 percent

of the estimations report a σ̂ that is significant at the ten percent level indicating the appropriateness of

the nested logit structure. Row 9 reports that the quality estimates are precisely estimated, which is not

surprising since these estimates are the sum of two fixed effects plus a residual.18

4.1 Factor Endowments and Quality Specialization

The inferred qualities offer support for previous studies that have found, using prices to proxy for quality,

that more capital- and skill-intensive countries export higher quality varieties. The relationship between

export quality inferred using the method above and level of development is assessed by regressing variety1712 percent of the SITC industries record imports in multiple units. For these industries, the products of the majority unit

are kept which comprise about 80 percent of the observations within a multiple-unit industry.18The standard errors are obtained by simulating draws from the asymptotic distribution of the estimated parameters.

13

quality on GDP per capita

λckt = αkt + β ln Yct + νckt, (16)

where λckt is the estimated quality of country c’s export in product k at time t and Yct is country c’s

per capita GDP. The inclusion of a product-year dummy, αkt, indicates that the regression considers the

cross-sectional relationship between quality and income within products. Table 4 reports that the coefficient

on exporter income is positive and significant. Richer countries, on average, export higher quality varieties

within products.

Columns 2 and 3 re-run (16) using capital-labor ratios and the fraction of a country’s workforce

with tertiary education. The coefficient on capital-labor endowment is positive and significant, so more

capital intensive countries also tend to export higher quality varieties within products. The coefficient on

the education variable is positive, but not statistically significant, but the precision may be low due to lack

of data. The three regressions in Table 4 support previous work that has argued, using unit values, that

developed countries export higher quality varieties (e.g., see Schott (2004) or Hallak (2006)).

4.2 Quality Ladders

Regression (16) provides evidence that richer countries sit atop the quality ladder within products. As in the

theoretical model, I define a product’s quality ladder as the range of the estimated qualities within a product.

One concern is that the quality ladder may be changing over time as countries increase R&D expenditure, in

the case of developed countries, or gain access to improved technology, in the case of developing countries.

In order to minimize the potential for an endogenous quality ladder, I rely on a product’s quality ladder

measured in the first year (indexed by a 0) the product appears in the sample19

Ladderk0 = λmaxk0 − λmin

k0 . (17)

Fixing a product’s quality ladder may be problematic if initially “short” ladders become “long” or vice

versa. However, there is a strong persistence in a product’s ladder length over the sample. The correlation

coefficient between a product’s initial ladder length and its end of sample length is 0.84. This persistent

correlation indicates that the scope for quality differentiation is an intrinsic feature of products.

In markets characterized by long quality ladders, there is a relatively higher gradient between prices

and qualities, as measured here. Recall that in a vertical product market, prices and quality are isomorphic

since consumers agree on the rankings of goods. The mapping between prices and quality is less clear in hor-

izontal markets because idiosyncratic preferences influence purchasing decisions. The following specification19The main results of the paper actually rely on the inter-decile range which is more robust to outliers than the range. However,

the sensitivity checks reveal that the results are robust to alternative measures including the full range, the inter-quartile range

and the standard deviation of qualities within a product (see below).

14

assesses the relationship between the inferred qualities and prices across products of varying ladder lengths

ln pckt = αkt + β1λckt + β2 (λckt × ln Ladderk0) + νckt. (18)

Product-year fixed effects are denoted by αkt, and pckt is the unit value of country c’s export in product k

at time t. Table 5 reports that the interaction coefficient, β2, is positive and significant.20 So in markets

characterized by longer quality ladders, the gradient between prices and qualities is relatively higher. This is

consistent with the vertical market assumption frequently made in the literature. However, this correlation

weakens as the ladder length declines implying that prices may be imperfect proxies for quality in short-ladder

markets.

Regression (18) indicates that the average consumer does not attach a high valuation to expensive

imports in short-ladder products. For example, the estimated qualities reveal that while Canadian footwear

is 27 percent more expensive than average imported footwear, it has a lower than average estimated quality.

Horizontal attributes explain why these expensive, but low quality, Canadian shoes are purchased. A fraction

of consumers purchase Canadian footwear not because of a high λ but because this fraction obtains a high logit

draw for Canadian shoes. The average U.S. consumer, however, attaches a low valuation to these Canadian

shoes indicating that these imports do not offer vertical attributes to justify its high price. Inferring quality

from prices alone would instead attach a high quality rank for Canadian footwear.

Two graphs further illustrate this point. Figure 1 plots the relationship between quantities, unit

values and the estimated qualities for two products: “Transmission Receivers Exceeding 400 MHZ” (HS

8525203080) and “Footwear with Plastic Soles, Leather Uppers” (HS 6403999065). The figures are ordered

by unit values, which also roughly correspond to exporter per capita GDP. For transmission receivers (top

panel), unit values and quality are positively correlated, indicating that the average consumer assigns a higher

valuation to more expensive varieties. For this product, it appears that the vertical market assumption is

tenable.

The bottom panel plots leather shoes. Here, exporters of expensive varieties, like Belgium, are

associated with relatively low quality. The reason lies in the export quantities (square dots). Belgium has

a very low market share, even conditioning on its price. Taking into account Belgium’s market share and

export price, the quality estimates indicate that the average consumer attaches a low valuation to Belgian

leather shoes. On the other hand, France exported the second most expensive variety in this HS classification

and obtained a relatively high market share given its price. It is therefore assigned a high quality estimate.

China’s exceptionally large market share, conditional on its price, results in the highest export quality for

this HS code. Consistent with casual evidence, footwear exported by Spain, Italy and Germany are of

relatively high quality: these countries’ footwear are expensive but secure high market shares. These two20Note that the negative β1 coefficient is a consequence of how quality is defined (see equation (14)): conditional on market

shares, price and the estimated quality measures are positively correlated.

15

figures therefore suggest that the vertical market assumption frequently invoked in the trade literature may

be more reasonable for some product markets rather than others, and indeed these figures represent the key

intuition developed in this paper.

It is important that the quality ladders are constructed from not only from developing country

imports, but also imports from highly developed countries like Japan, Germany and Canada. The ladders

are therefore likely to represent the true quality frontier of a product. But while most countries export

apparel and footwear products, there is a negative correlation between the average number of varieties

within an HS code and industry capital intensity. Based on equation (14), a country has zero market share

if consumers assign a negative infinite consumer valuation (or if the price is infinite). In the next section,

I demonstrate that quality ladder lengths are positively correlated with industry capital intensity. Putting

these two findings together indicates that the selection bias that occurs because not all countries export more

capital-intensive products will underestimate the quality ladder for these products. As a result, accounting

for the selection bias (e.g., see Helpman et al. (2006)) is not a major concern since the selection bias works

against the results below. In other words, the quality ladders are underestimated for the markets that are

least affected by low-wage competition.

5 Long and Short Quality Ladders

5.1 Quality Ladders and U.S. Manufacturing Employment

This section examines the vulnerability hypothesis outlined in Section 2 by linking the impact of import

penetration on U.S. manufacturing employment with industry ladder lengths. Since U.S. employment data

is unavailable at the 10-digit HS level, I construct an SIC industry quality ladder, IndLadderm, to match to

industry-level U.S. employment data. The industry ladder is defined as the weighted average of the (initial

year) product ladders within the four-digit SIC industry:

IndLadderm =Km∑

k=1

w0kLadderk0, (19)

where Km denotes the number of ten-digit HS products in SIC industry m, Ladderk0 is the product’s baseline

ladder (defined in (17)). The product’s weight within an SIC is obtained from its initial year as well. Again,

by relying on initial year values, the industry ladder becomes a time invariant measure.

The industry ladders are matched to the NBER manufacturing database from 1989-96 (Bartelsman

et al., 1996). Summary statistics for the quality ladders are shown in the final column of Table 2. Table 6

provides a decomposition of the industry ladder by observable industry characteristics: skill intensity, capital

intensity and total factor productivity.21 All three coefficients are positive, and the correlation with capital21Skill intensity is measured as the ratio of non-production to production workers. Capital intensity is the ratio of capital

16

intensity is statistically significant. However, notice that the R-squared is extremely low suggesting that

most of the variation in the quality ladder cannot be explained by these variables.

Following Bernard et al. (2006), I link employment outcomes with two measures of import penetra-

tion: imports originating from countries with less than 5 percent of U.S. per capita GDP (LWPEN) and

the rest of the world (OTHPEN). Total import penetration is defined as Imt/(Imt + Qmt −Xmt), where

Imt is the value of imports in four-digit SIC industry m at time t, Qmt is the industry’s domestic production

and Xmt represents U.S. exports. LWPEN is the product of total import penetration and the value share

of imports originating from low-wage countries

LWPENmt =I lowmt

Imt + Qmt −Xmt. (20)

OTHPEN is defined analogously as

OTHPENmt =Imt − I low

mt

Imt + Qmt −Xmt. (21)

The following specification regresses industry employment outcomes on the industry quality ladder

and the import penetration measures

ln Empmt = αm + αt + β1OTHPENmt + β2LWPENmt + β3 (LWPENmt × ln IndLadderm) + νmt. (22)

Note that the specification includes both industry (αm) and year (αt) fixed effects. If the data are consistent

with the quality ladder hypothesis outlined in Section 2, the results should find β1, β2 < 0; higher import

penetration is negatively correlated with industry employment. The coefficient of interest is the interac-

tion between LWPEN and IndLadder, which measures the differential impact of low-wage penetration

on employment across industries of varying ladder lengths. The vulnerability hypothesis predicts β3 > 0;

long-ladder industries with high exposure to low-wage countries suffer smaller employment declines.22

Column one of Table 7 reports the baseline results. The coefficients are statistically significant

and have the predicted signs. Import penetration negatively affects employment, and imports from low-

wage countries have a larger impact than imports originating from the remaining (richer) countries. The

interaction coefficient is positive and precisely estimated, supporting the model’s prediction that vulnerability

to low-wage penetration declines in industries with longer quality ladders.

The point estimates are also economically significant. If low-wage penetration increases by ten

percentage points, employment in an average ladder industry declines by 6.5 percent. In contrast, low-wage

penetration is associated with only a 1.6 percent employment loss in a long-ladder industry (one standard

deviation above the mean). For a specific example, if LWPEN were to increase by ten percentage points in

stock to total employment. Total factor productivity is obtained from a five-factor model (Bartelsman et al., 1996).22Note that the predictions for the signs of β1, β2, β3 are opposite from the model because in the model, an increase in wages

is analogous to a decline in LWPEN .

17

footwear (SIC 314), employment would fall 15 percent more than in household audio and video equipment

(SIC 365), an industry characterized by a three times longer quality ladder.

In column two, I include a ladder-OTHPEN interaction to determine if the effects of imports

originating from more-advanced countries are also dampened in long ladders. This interaction is statistically

significant, but its economic magnitude is smaller than the ladder-LWPEN interaction term.

Given that the quality ladder is positively correlated with industry capital intensity, (see Table 6),

one concern is that the quality ladder simply proxies this variable. If this is true, then the results in

columns one and two simply confirm the findings of previous studies arguing that more capital-intensive

industries are less susceptible to import competition (e.g., see Bernard, Jensen, and Schott (2006)). To

address this concern, I include the interaction of capital-intensity with LWPEN in column three.23 More

capital-intensive industries are less vulnerable to low-wage imports, as expected, but the quality ladder

interaction remains positive and significant. Moreover, the magnitudes of the point estimates are about the

same. Employment in a short-ladder is predicted to fall 4.6 percent more, ceteris paribus, than its long-

ladder counterpart. Likewise, a low capital-intensive industry would contract 7.5 percent more than a highly

intensive industry due to low-wage competition.

While using the baseline ladder and factor intensities mitigate endogeneity concerns, import pene-

tration may be endogenous. For instance, international trade may be filling a void created by a decline in

domestic industries caused by other factors, such as structural changes in the economy. The simultaneity

would bias the import penetration coefficients downward in (22). I therefore instrument the penetration

measures with industry-year weighted averages of exchange rates, where the weights are the country’s share

of industry value in 1989, for low-wage countries and the rest of the world. Tariffs and freight rates also

serve as instruments. These two instruments are constructed by dividing total duties and freight costs for

each set of countries divided by their total import value by industry-year.

The fourth column of Table 7 presents the baseline IV specification. The first column shows the

baseline specification. Instrumenting actually causes the coefficient on LWPEN to increase in magnitude,

which suggests measurement error in the variable.24 The quality ladder now generates an even larger impact

of competition on employment. For example, a ten percentage point increase in LWPEN leads to a 9 percent

employment decline in a short-ladder industry (one standard deviation below the mean) compared to a 15.6

percent employment gain in the average industry. The result remains robust to the OTHPEN − Ladder

interaction in column 5. Finally, column 6 includes the interaction of low-wage penetration with industry

capital intensity and the magnitude of the estimates are similar, as was the case with the OLS regression in

column 3. Employment in a short-ladder is predicted to fall 15.7 percent more, ceteris paribus, than its long-23Since this variable is itself endogenous, the regression fixes the capital intensity at its 1989 values. This also means that

the levels are not identified because of the industry fixed effects.24Bernard et al. (2006) also find that instrumenting import penetration causes the magnitude of the coefficients to increase.

18

ladder counterpart while a low capital-intensive industry would contract 16.9 percent more than a highly

capital-intensive industry if low-wage penetration were to rise ten percentage points. This indicates that

even industries with similar observable characteristics may exhibit heterogenous impacts from international

trade because of inherent differences in vertical specialization.

The point estimates suggest large impacts, but they are consistent with a argument emphasized by

Leamer (2000): even low import volumes can have a significant impact on U.S. firms if international trade

equalizes product prices. The results indicate this argument is particularly salient for short-ladder products.

Indeed, the extent to which domestic goods overlap with foreign goods, and the source of the foreign imports,

is precisely what determines which industries are vulnerable to competition in the framework here. The

magnitude of the employment effects are also consistent with Bernard et al. (2006), whose conservative

estimates indicate that a ten percentage point increase in LWPEN raises the probability of U.S. plant

death by 17 percent. Moreover, the raw data reveal large correlations between employment outcomes and

rising low-wage import penetration. For example, the household slipper industry’s (SIC 3142) quality ladder

is half the average and between 1989 and 1996, employment fell more than 50 percent while low-wage

penetration simultaneously rose 25 percentage points. Import competition therefore can have large impacts

on domestic firms in short-ladder industries, particularly those at the competitive fringe.

Finally, Table 8 reruns (22) with industry output as the dependent variable to show that employment

outcomes are not simply an artifact of U.S. firms substituting labor with capital. This table is also a closer

link to the model which linked changes in the nature of competition to output. The table shows that the

Ladder interaction is positive and significant across all six specifications and the magnitudes are comparable

with the employment regressions. For example, using the point estimates in column one, the impact of

low-wage penetration on output growth in a short- versus long-ladder is 7 percent more. Thus, the results

offer strong evidence that long-ladder industries contract less than short-ladder industries given the same

level of low-wage import penetration.

5.1.1 Robustness Checks

I perform a number of robustness exercises to check the sensitivity of the results. The first check re-runs

the IV specifications using two-digit SIC-year fixed effects to control for time-varying unobservables at the

sector-level. This flexible specification, for example, controls for technological changes (at the two-digit

level) that may be correlated with the initial quality ladder. The regressions for employment and output are

reported in Table 9. The magnitude of the coefficients declines, not surprisingly, but the interaction remains

statistically significant.

Table 10 reports additional sensitivity checks by re-running (22) using alternative measures of the

industry quality ladder. Each row reports the LWPEN -Ladder coefficient from the OLS and IV regressions.

19

In the first row, the quality ladder is constructed by excluding any intermediate HS product.25 While crude,

this is another approach to control for the potential transfer pricing arrangements of intermediate products.

The results are not sensitive to excluding this set of HS products.

Constructing the quality ladders hinges on the disaggregate detail of U.S. import data. One concern

might be that the ladder lengths simply reflect aggregation differences if products in some industries are

defined more coarsely than others. Another worry could be that the product-level ladders just proxy for

the number of countries exporting that product code. To ensure that the results are not sensitive to these

concerns, rows 2 and 3 re-run the employment regressions using “count” definitions of the ladder. The second

row defines the product-level ladder as the number of countries (i.e., varieties) within the product and then

aggregates to the industry level according to (19). Row 3 counts the number of products within the four-digit

industry code and uses that as the industry-level ladder. Both measures proxy potential differences in the

coarseness of product definitions across industries. However, the OLS and IV coefficients are imprecisely

measured. This provides evidence that the baseline results are not driven by the coarseness of data.

Regression (18) provides evidence that prices may not be equivalent to quality in some markets,

particularly those with a high degree of horizontal product differentiation. However, theoretical and empirical

models typically use unit values as the a proxy for quality. Row 4 constructs a quality ladder from the

dispersion of unit values within products. While the OLS coefficient is not significant at conventional levels,

the IV coefficient is significant at conventional levels. The IV result is also robust to including the capital-

intensity interaction (e.g., see column 6 of Table 7). While acknowledging the caveat that unit values may

be imperfect proxies for quality, as discussed extensively above, this finding offers further support for the

quality ladder hypothesis.

Row 5 of Table 10 takes the opposite approach and constructs the ladder using within-product

market share dispersion. While the OLS coefficient is statistically significant, the IV coefficient is very

imprecisely estimated. This is not surprising; high quality is not assigned to products simply with high

market shares, but rather high market shares conditional on price.

Row 6 defines quality exclusive of the residual from the estimating equation in (14), λckt = λ1ck+λ2t,

and then constructs the industry ladder using (17) and (19). This measure addresses potential concerns that

the residual term (∆λ3ckt) may be capturing factors other than quality. However, the table shows that the

results are robust to defining quality without this term.

The definition of quality in (15) is measured in utils relative to the outside good. The estimated

price coefficient acts as an exchange rate between dollars and utils so quality can be converted into dollar-

denominated value by dividing by the price coefficient. Row 7 indicates that the results are not sensitive to25Intermediate goods are identified as any HS code that contains the phrases “parts”, “prts”, “pts” or “component” in its

description.

20

using a dollar-denominated quality measure.

Row 8 of Table 10 constructs the quality ladder for quality estimates obtained from specification

(14) where population is replaced by GDP as the proxy for hidden varieties. The results are not sensitive to

using this alternative proxy, either.

Row 9 adds interactions of skill intensity and TFP (using 1989 values) with low-wage penetration.

The results are not sensitive to the inclusion of these additional interactions. This is important because

it shows that the baseline results are not simply driven by heterogenous impacts across standard industry

characteristics such as skill, capital intensity or TFP.26

The next sensitivity check uses the HS-level elasticities of substitution estimated by Broda and

Weinstein (2006). These elasticities measure the degree to which horizontal varieties, within HS codes, are

substitutable. I aggregate their elasticities up to the industry level and report the Broda-Weinstein ladder

coefficients in row 10. The OLS coefficient is significant at the 10 percent level and has an intuitive sign:

industries with higher substitutability experience relatively more smaller employment growth. However, the

IV coefficient is not statistically significant. Given that the elasticities of substitution capture the degree of

horizontal differentiation, this finding is consistent with the theoretical model above that emphasizes vertical

specialization in reducing exposure to low-wage competition over horizontal differentiation.

The unit values used in the baseline regressions are inclusive of tariff charges. One potential caveat

is that the unit values exclude information on non-tariff barriers, such as voluntary export restraints or

quotas. The major non-tariff barriers during this period were quotas imposed on textile and apparel imports

under the Multifiber Arrangement (MFA) and its successor, the Agreement on Textile and Clothing (ATC).

I therefore re-run the regressions by constructing quality ladders that exclude the HS products that were

subject to quotas under the MFA/ATC. These HS products are obtained from Brambilla, Khandelwal, and

Schott (2008). Row 11 indicates that the results are not sensitive to excluding products that were subject

to import quotas.

Finally, the remaining rows construct quality ladders using alternative measures of dispersion: range,

inter-quartile range and the standard deviation of the estimated qualities. All coefficients in both the OLS

and IV regressions are positive and statistically significant, so the quality ladder measure is not sensitive to

alternative metrics for measuring quality dispersion.

In short, Table 10 indicates that the heterogenous impact of low-wage penetration across industries

of varying quality ladders is robust to several critiques and alternative measure of industries’ quality ladders.26Skill intensity is measured as the ratio of non-production to production workers. Capital intensity is the ratio of capital

stock to total employment. Total factor productivity is obtained from a five-factor model (Bartelsman et al., 1996).

21

6 Conclusion

Product specialization can weaken the convergence in goods and factor prices predicted to result from inter-

national trade. This paper builds on previous literature that has argued that specialization occurs within

products along the quality dimension. Products, however, vary in their scope for quality differentiation.

This paper offers evidence that variation in quality ladders has important implications for how international

trade affects developed countries. If developed countries are unable to exploit comparative-advantage factors

to manufacture vertically superior goods, employment and output in these products is likely to shift towards

lower cost countries. I find support for this theory by matching the quality ladders to U.S. industry em-

ployment outcomes resulting from international trade. Consistent with the model, the impact of low-wage

import penetration on employment varies inversely with the ladder length.

The analysis fixes the quality ladder length which is a reasonable assumption in the short- to medium-

run. In the long run, the ladder is, of course, endogenous as firms respond to competition through investment

and innovation to lengthen the quality ladder. This paper ignores the product cycling through which devel-

oped countries innovate new products that eventually migrate to the South.27 Addressing the dynamics of

import competition and innovation activity, or other means of protection, such as lobbying for trade policy,

are extensions to this research that will continue to draw from developments in the international trade and

industrial organization literatures.

In addition to the effects on firm survival, quality upgrading in response to import competition

may have implications for rising income inequality within developed and developing countries. Climbing the

existing quality ladder, or investing in innovative resources to lengthen the quality ladder, presumably raises

the demand for skilled labor. If so, this would suggest an additional channel through which globalization

increases income inequality in addition to other factors, such as the offshoring of manufacturing activity to

developing countries (Feenstra and Hanson, 1999). This paper offers an alternative approach to estimating

product quality using both prices and market share information. This procedure may prove useful in fur-

thering our understanding of the firm-level mechanisms of quality specialization resulting from more intense

import competition, and its subsequent impact on workers.

References

Aghion, P. and P. Howitt (1992). A model of growth through creative destruction. Econometrica 60, 323–351.

Anderson, S., A. De Palma, and J. Thisse (1987). The CES is a discrete choice model? Economic Letters 24,

139–140.27See Vernon (1966), Grossman and Helpman (1991), and Aghion and Howitt (1992).

22

Anderson, S., A. de Palma, and J. Thisse (1992). Discrete Choice Theory of Product Differentiation. Cam-

bridge: MIT Press.

Bartelsman, E., R. Becker, and W. Gray (1996). The NBER-CES manufacturing industry database. NBER

Technical Working Paper No. 205.

Bernard, A., J. Jensen, and P. Schott (2006). Survival of the best fit: Exposure to low-wage countries and

the (uneven) growth of U.S. manufacturing. Journal of International Economics 68, 219–237.

Bernard, A., S. Redding, and P. Schott (2007). Comparative advantage and heterogeneous firms. Review of

Economics Studies 74, 31–66.

Berry, S. (1994). Estimating discrete-choice models of product differentiation. The RAND Journal of

Economics 25, 242–262.

Berry, S., J. Levinsohn, and A. Pakes (1995). Automobile prices in market equilibrium. Econometrica 63,

841–890.

Brambilla, I., A. Khandelwal, and P. Schott (2008). China’s experience under the Multifiber Arrangement

and Agreement on Textile and Clothing. Columbia University, Mimeo.

Broda, C. and D. Weinstein (2006). Globalization and the gains from variety. Quarterly Journal of Eco-

nomics 121, 541–585.

Brooks, E. (2003). Why don’t firms export more? Product quality and Colombian plants. UC Santa Cruz,

Mimeo.

Cardell, N. S. (1997). Variance components structures for the extreme-value and logistic distributions with

application to models of heterogeneity. Econometric Theory 13, 185–213.

Davis, D. and D. Weinstein (2001). An account of global factor trade. American Economic Review 91,

1423–1454.

Feenstra, R. (1994). New product varities and the measurement of international prices. American Economic

Review 84, 157–177.

Feenstra, R. and G. Hanson (1999). The impact of outsourcing and high-technology capital on wages:

estimates for the United States, 1979-1990. Quarterly Journal of Economics 114, 907–940.

Feenstra, R., J. Romalis, and P. Schott (2002). U.S. imports, exports, and tariff data, 1989-2001. NBER

Working Paper 9387.

23

Flam, H. and E. Helpman (1987). Vertical product differentiation and north-south trade. American Economic

Review 77, 810–822.

Freeman, R. and L. Katz (1991). Industrial wage and employment determination in an open economy. In

J. Abowd and R. Freeman (Eds.), Immigration, Trade and Labor Market. Chicago: University of Chicago

Press.

General Accounting Office (1995). U.S. imports: Unit values vary widely for identically classified commodi-

ties. Report GAO/GGD-95-90.

Goldberg, P. and N. Pavcnik (2007). Distributional effects of globalization in developing countries. Journal

of Economic Literature 45, 39–82.

Grossman, G. and E. Helpman (1991). Quality ladders and product cycles. Quarterly Journal of Eco-

nomics 106, 557–586.

Hallak, J. and P. Schott (2007). Estimating cross-country differences in product quality. Yale University,

Mimeo.

Hallak, J. C. (2006). Product quality and the direction of trade. Journal of International Economics 68,

238–265.

Helpman, E., M. Melitz, and Y. Rubinstein (2006). Trading partners and trading volumes. Harvard Univer-

sity, Mimeo.

Hidalgo, C., B. Klinger, A. Barabasi, and R. Hausmann (2007). The product space conditions the develop-

ment of nations. Science 317, 482–487.

Irwin, D. and N. Pavcnik (2004). Airbus versus boeing revisited: International competition in the aircraft

market. Journal of International Economics 64, 223–245.

Klette, T. J. and S. Kortum (2004). Innovating firms and aggregate innovation. Journal of Political Econ-

omy 112, 986–1018.

Krugman, P. (1980). Scale economies, product differentiation and the pattern of trade. American Economic

Review 70, 950–959.

Kugler, M. and E. Verhoogen (2008). Product quality at the plant level: Plant size, exports, output prices

and input prices in colombia. Mimeo, Columbia University.

Leamer, E. (1987). Paths of development in the three-factor, n-good general equilibrium model. Journal of

Political Economy 95, 961–999.

24

Leamer, E. (2000). What’s the use of factor contents? Journal of International Economics 50, 17–49.

Leamer, E. (2006). A flat world, a level playing field, a small world after all, or none of the above? UCLA,

Mimeo.

Melitz, M. (2003). The impact of trade on intra-industry reallocations and aggregate industry productivity.

Econometrica 71, 1695–1725.

Rauch, J. (1999). Networks versus markets in international trade. Journal of International Economics 48,

7–35.

Revenga, A. (1992). Exporting jobs? The impact of import competition on employment and wages in U.S.

manufacturing. Quarterly Journal of Economics 107, 255–284.

Rodrik, D. (2006). What’s so special about China’s exports? NBER Working Paper 11947.

Sachs, J. and H. Shatz (1994). Trade and jobs in U.S. manufacturing. Brookings Papers on Economic

Activity 1994, 1–69.

Schott, P. (2003). One size fits all? Hecksher-Ohlin specialization in global production. American Economic

Review 93, 686–708.

Schott, P. (2004). Across-product versus within-product specialization in international trade. Quarterly

Journal of Economics 119, 647–678.

Schott, P. (2008). The relative sophistication of Chinese exports. Economic Policy 23, 5–49.

Sutton, J. (1991). Price Competition, Advertising, and the Evolution of Concentration. Cambridge: The

MIT Press.

Sutton, J. (1998). Technology and Market Structure: Theory and History. Cambridge: The MIT Press.

Verhoogen, E. (2008). Trade, quality upgrading, and wage inequality in the Mexican manufacturing sector.

Quarterly Journal of Economics 123, 489–530.

Vernon, R. (1966). International investment and international trade in the product cycle. Quarterly Journal

of Economics 80, 190–207.

25

7 Tables

Table 1: Low-wage Countries

Afghanistan Chad Haiti Niger

Albania China India Pakistan

Angola Congo Kenya Rwanda

Armenia Equitorial Guinea Lao PDR Samoa

Azerbaijan Ethiopia Madagascar Sierra Leone

Bangladesh Gambia Malawi Sri Lanka

Benin Georgia Mali Sudan

Burkina Faso Ghana Mauritania Togo

Burundi Guinea Moldova Uganda

Cambodia Guinea-Bissau Mozambique Vietnam

Central African Republic Guyana Nepal Yemen

Notes: Notes: The table provides the list of low-wage countries used in the paper. Low-wage

countries are defined as countries with a less than 5 percent of US per capita GDP. Source:

Bernard et al (2006).

26

Table 2: Summary Statistics

Sector (SIC-2)

Industry

(SITC-5)

Products

(HS-10) Varieties Countries

Varieties/

Product Share

Average

GDP

Ladder

Length

20 Food 8 37 588 69 15.89 0.00 16,828 1.86

22 Textile 85 1,639 23,658 128 14.43 0.01 13,175 2.42

23 Apparel 68 2,539 52,060 139 20.50 0.09 6,996 2.14

24 Lumber 20 262 3,931 110 15.00 0.01 12,768 1.87

25 Furniture 5 72 2,496 104 34.67 0.01 11,673 1.48

26 Paper 38 216 4,284 104 19.83 0.02 19,737 1.86

27 Printing 16 55 2,237 114 40.67 0.01 17,404 1.00

28 Chemicals 231 2,557 33,977 133 13.29 0.07 20,038 2.43

29 Petroleum 7 21 448 81 21.33 0.00 10,758 2.72

30 Rubber & Plastic 45 514 9,234 112 17.96 0.02 13,990 2.28

31 Leather 17 403 9,295 122 23.06 0.03 5,931 1.68

32 Stone & Ceramic 57 357 8,698 122 24.36 0.01 14,976 1.99

33 Primary Metal 98 1,372 21,605 112 15.75 0.04 16,696 2.25

34 Fabricated Metal 78 599 15,372 126 25.66 0.03 17,224 1.52

35 Industrial Machinery 169 1,664 32,021 130 19.24 0.15 20,339 2.35

36 Electronic 100 1,334 29,627 135 22.21 0.18 15,134 1.93

37 Transportation 43 372 5,940 112 15.97 0.25 22,943 2.19

38 Instruments 60 715 10,785 111 15.08 0.03 21,624 2.37

39 Miscellaneous 76 375 8,041 132 21.44 0.04 10,690 2.11

Notes: The table provides summary statistics for the two-digit SIC (1987 revision) sectors. Column four

reports total import varieties (country-product pairs in each SIC sector between 1989 and 2001). Column

seven reports the weighted average of exporter per capita GDP. Column eight reports the (log) weighted

average industry ladder (see equation (19)). Source: Feenstra et al (2002), World Development Indicators

and author's calculations.

Table 3: Quality Estimation Results

Statistic Mean Median 1st Quarter 3rd Quartile

OLS price coefficient -0.046 -0.003 -0.020 -0.0002

IV price coefficient -0.080 -0.007 -0.063 -0.0002

*Own price elasticity -1.54 -0.63 -1.57 -0.22

Overidentifying restrictions p-value 0.236 0.123 0.016 0.398

1st stage F-stat p-value, price 0.030 0.000 0.000 0.004

1st stage F-stat p-value, nest share 0.031 0.000 0.000 0.005

Coefficient on conditional market share 0.42 0.45 0.08 0.84

Coefficient on population 0.70 0.47 -1.43 3.28

T-statistic of quality estimates 11.22 3.81 1.31 10.64

R-squared 0.24 0.14 0.04 0.41

Observations per estimation 1,091 467 211 1,085

Estimations with stat. sig. price coefficient

Observations with stat. sig. price coefficient

Total Estimations

Total Number of Products

Total Observations

1,059

0.70

0.80

15,103

1,232,764

Notes: Table reports estimation statistics of running equation (14) separately for each of the

1,059 manufacturing sectors. Each regression clusters standard errors by exporting country. *

denotes that the elasticity is computed conditional on a negative price coefficient and � (0,1].

27

Table 4: Quality and Factor Endowments

Regressors (1) (2) (3)

Log (PCGDPct) 0.523 ***

0.128

Log (KLct) 0.521 ***

0.123

Log (Educationct) 0.231 *

0.130

Product x Year FEs yes yes yes

R-squared 0.22 0.22 0.34

Observations 1,232,252 1,162,803 511,749

Qualityckt

Notes: Table regresses the quality estimates on (log) per capita GDP, (log)

capital-labor ratios and percentage of workforce with tertiary education.

Regressions include product-year fixed effects. Robust standard errors are

clustered by exporting country. Significance levels: *** .01 ** .05 * .1.

Source: World Bank's World Development Indicators and Penn World Table

(6.1).

28

Table 5: Relationship between Quality and Price

Regressors

Qualityckt -0.026 ***

0.010

Qualityckt x Log (Ladder0k) 0.006 ***

0.002

Product x Year FEs yes

R-squared 0.87

Observations 1,232,764

ln(priceckt)

Notes: Table regresses variety quality on (log) price and its

interaction with the product's baseline quality ladder. Robust

standard errors are clustered by exporting country. Significance

levels: * .1 ** .05 *** .01.

29

Table 6: Quality Ladder Decomposition

Regressors

Log (K/Lm) 0.141 ***

0.051

Log (Skillm) 0.002

0.067

Log (TFPm) 0.423

0.600

Constant 1.516 ***

0.231

R-squared 0.03

Observations 325

Log (IndLadderm)

Notes: Table regresses the industry quality ladder (defined in

(19)) on four-digit SIC factor intensities: capital intensity, skill

intensity and total factor productivity. Capital intensity is the ratio

of capital stock to total employment. Skill intensity is measured

as the ratio of non-production to production workers. Total factor

productivity is obtained from a five-factor model (see Bartelsman

et al (1996)). Factor intensities measured at 1989 values.

30

Table 7: Employment Regressions

Regressors (1) (2) (3) (4) (5) (6)

OTHPENmt -0.475 *** -0.990 *** -0.457 *** -0.710 0.561 -0.627

0.161 0.356 0.135 0.783 1.809 0.466

Log (IndLadderm) x OTHPENmt 0.246 ** -0.719

0.111 0.995

LWPENmt -2.001 *** -2.114 *** -4.581 *** -5.239 *** -6.456 *** -9.773 ***

0.277 0.312 1.116 1.365 2.173 2.125

Log (IndLadderm) x LWPENmt 0.653 *** 0.684 *** 0.609 ** 3.295 *** 4.344 ** 2.093 ***

0.195 0.205 0.280 1.146 1.913 0.605

Log (K/Lm) x LWPENmt 0.881 ** 1.983 ***

0.371 0.759

Overidentification p-value - - - 0.25 0.60 0.35

Kleibergen-Paap F statistic 1.2 3.5 11.6

Industry, Year FEs yes yes yes yes yes yes

R-squared 0.11 0.12 0.12 0.002 0.000 0.013

Observations 2,585 2,585 2,585 2,585 2,585 2,585

OLS IV

Log (Employmentmt)

Notes: The dependent variable for each regression is the four-digit SIC industry (log) employment. The first column regresses

employment on import penetration from the rest of the world (OTHPEN ), low-wage import penetration (LWPEN ) and the

interaction of LWPEN with the industry quality ladder. Column 2 includes the OTHPEN -ladder interaction. Column 3 includes

the interaction of LWPEN with industry capital intensity (in 1989). Columns 4-6 present the IV results. The instruments are

weighted average tariff rates, exchange rates and freight rates for low-wage countries and the rest of the world. Robust

standard errors are clustered at the two digit SIC. Significance levels: * .1 ** .05 *** .01.

31

Table 8: Output Regressions

Regressors (1) (2) (3) (4) (5) (6)

OTHPENmt -0.494 ** -0.956 * -0.471 ** 0.705 4.091 ** 0.585 **

0.207 0.547 0.185 1.110 1.685 0.259

Log (IndLadderm) x OTHPENmt 0.221 -1.917

0.236 1.406

LWPENmt -2.358 *** -2.459 *** -5.603 *** -5.986 *** -9.229 *** -13.867 ***

0.440 0.474 1.884 2.055 3.413 2.739

Log (IndLadderm) x LWPENmt 0.882 ** 0.910 ** 0.828 * 4.537 ** 7.332 ** 2.715 **

0.378 0.378 0.472 1.792 3.166 1.086

Log (K/Lm) x LWPENmt 1.108 3.296 ***

0.781 1.226

Overidentification p-value - - - 0.51 0.45 0.66

Kleibergen-Paap F-statistic 1.2 3.5 11.6

Industry, Year FEs yes yes yes yes yes yes

R-squared 0.19 0.19 0.20 0.04 0.04 0.03

Observations 2,585 2,585 2,585 2,585 2,585 2,585

OLS IV

Log (Outputmt)

Notes: The dependent variable for each regression is the four-digit SIC industry (log) output. The first column regresses

employment on import penetration from the rest of the world (OTHPEN), low-wage import penetration (LWPEN) and the

interaction of LWPEN with the industry quality ladder. Column 2 includes the OTHPEN-ladder interaction. Column 3 includes

the interaction of LWPEN with industry capital intensity (in 1989). Columns 4-6 present the IV results. The instruments are

weighted average tariff rates, exchange rates and freight rates for low-wage countries and the rest of the world. Robust

standard errors are clustered at the two digit SIC. Significance levels: * .1 ** .05 *** .01.

32

Table 9: Employment Regressions, SIC-Year Fixed Effects: IV

Regressors (1) (2) (3) (4) (5) (6)

OTHPENmt 0.143 -0.391 -0.241 1.085 2.126 0.458

1.168 2.772 0.494 1.269 3.218 0.324

Log (IndLadderm) x OTHPENmt 0.242 -0.471

0.930 1.212

LWPENmt -3.276 *** -2.967 ** -8.370 * -2.921 *** -3.522 ** -10.719 ***

1.269 1.472 4.391 1.074 1.751 2.079

Log (IndLadderm) x LWPENmt 2.084 *** 1.825 1.622 *** 2.477 *** 2.983 * 1.574 **

0.724 1.483 0.591 0.811 1.772 0.796

Log (K/Lm) x LWPENmt 1.692 2.488 ***

1.425 0.910

Overidentification p-value 0.23 0.45 0.43 0.48 0.35 0.32

Kleibergen-Paap F-statistic 1.78 1.47 4.67 1.78 1.47 4.67

Industry FEs yes yes yes yes yes yes

Two-Digit SIC x Year FEs yes yes yes yes yes yes

R-squared 0.001 0.0002 0.014 0.042 0.050 0.003

Obs. 2,585 2,585 2,585 2,585 2,585 2,585

IV - Log (Employmentm) IV - Log (Outputm)

Notes: Notes: The dependent variables for each regression in columns 1-3 and columns 4-6 are the four-digit SIC industry (log)

employment and output, respectively. Each regression includes two-digit SIC-year pair fixed effects. The first column regresses

employment on import penetration from the rest of the world (OTHPEN ), low-wage import penetration (LWPEN ) and the interaction

of LWPEN with the industry quality ladder. Column two includes the OTHPEN -ladder interaction. Column three presents the IV

results. The instruments are weighted average tariff rates, exchange rates and freight rates for low-wage countries and the rest of

the world. Columns 4-6 run analogous regressions with output as the dependent variable. Robust standard errors are clustered at

the two digit SIC. Significance levels: * .1 ** .05 *** .01.

33

Table 10: Employment Regressions, Robustness Checks

LWPEN -Ladder Interaction

Quality Ladder ex. Intermediates 0.632 *** 3.027 ***

0.192 1.082

Product Count 0.000 -0.005

0.001 0.006

Variety Count -0.427 0.757

0.350 0.915

Price 1.651 7.619 **

1.141 3.853

Market Share 11.970 *** 17.697

3.434 20.244

Quality Ladder w/out Residual 0.695 *** 3.033 ***

0.177 1.044

$-Denominated Quality Ladder 0.211 * 1.041 *

0.108 0.599

GDP-Quality Ladder 0.653 *** 3.295 ***

0.195 1.146

Skill, TFP 0.572 ** 1.762 ***

0.235 0.671

Broda-Weinstein Ladder -0.448 * 0.101

0.256 0.511

0.557 *** 2.637 ***

0.162 0.898

Range 0.603 *** 2.384 ***

0.186 0.646

Inter-quartile Range 0.665 *** 3.673 ***

0.172 1.348

Standard Deviation 0.732 *** 3.596 ***

0.223 1.276

Excluding Textile and Apparel Products subject to Quotas

under the ATC/MFA

IV

Regression Method

OLS

Notes: Table re-runs the employment regression in equation (14) and reports the industry

ladder interaction with LWPEN coefficient for alternative definitions of the quality ladder. Row

1 defines a quality ladder that excludes intermediate HS products. Row 2 constructs the

ladder from the number of countries (i.e., varieties) within the product and then aggregates to

the industry according to equation (14). Row 3 counts the number of products within the four-

digit industry code. Rows 4-5 construct the ladder from the dispersion of prices and market

shares within products, respectively. Row 6 define the ladder from quality estimates exclusive

of the residual, ckt = 1ck+ 2t. Row 7 defines the ladder from quality estimates that are

converted from utils to dollar values by dividing the quality estimate by its price coefficient.

Row 8 re-runs equation (19) but uses GDP instead of population to control for hidden

varieties. Row 9 adds interactions of low-wage penetration with industry skill and TFP

measures to columns 3 and 6 of Table 7, respectively. Row 10 constructs the quality ladder

from the HS-level elasticities of substitution estimated by Broda and Weinstein (2006). Row 11

re-contructs the industry ladder by excluding HS products that were subject to quotas under

the ATC/MFA. The remaining rows use the full range, inter-quartile range and standard

deviation of qualities within products. Robust standard errors are clustered at the two-digit SIC

for each regression. Significance levels: * .1 ** .05 *** .01.

34

Figure 1: Prices, Market Shares and Quality

−10

−5

05

1015

(Log

) P

rice/

Qua

ntity

, Qua

lity

PH

L

CH

N

HK

G

IDN

ZA

F

IRL

TH

A

MY

S

ME

X

JPN

SG

P

FIN

SW

E

IND

TA

I

BR

A

KO

R

AU

T

AR

G

CH

E

GB

R

NO

R

NZ

L

ISR

ISL

CA

N

DE

U

ITA

NLD

JOR

FR

A

DN

K

TZ

A

AU

S

VE

N

Country

Price Quantity Quality Quality Best Fit

HS 8525203080Transmission Receivers Exceeding 400 MHZ

−5

05

10(L

og)

Pric

e/Q

uant

ity, Q

ualit

y

AR

GA

LB IND

ZA

FU

KR

CH

NLK

AT

AI

HK

GR

OM

PO

LID

NB

GR

UR

YB

RA

CO

LV

NM

DO

MB

LRP

AN

ES

PG

BR

GR

CM

AC

BH

ST

UR

ME

XT

UN

CZ

RK

OR

PH

LS

VN

ITA

TH

AD

NK

MC

DS

VK

NZ

LS

EN

MY

SA

US

SLV

CH

EP

ER

SW

EP

RT

DE

UC

AN

SLE

HU

NV

EN

CY

PM

DA

MA

RIS

RN

LDH

RV

AU

TF

RA

BE

L

Country

Price Quantity Quality Quality Best Fit

HS 6403999065Footwear with Plastic Soles, Leather uppers

The graphs show the price, quantity and estimated quality (and 95 percent confidence interval) for

countries for HS 8525203080 (“Transmission Receivers Exceeding 400 MHZ”) and HS 6403999065

(“Footwear with Plastic Soles, Leather Uppers”) in 2001. Countries are ordered by unit value.

35