THE IMPACT OF POST-9/11 VISA POLICIES
ON TRAVEL TO THE UNITED STATES
Brent Neiman
Harvard University
Phillip Swagel
Department of the Treasury
June 2007
This paper examines the impact of post-9/11 changes in visa and security policy on business
and leisure travel to the United States. American businesses, tourism industry
representatives, and politicians pointed to changes in visa policies as being responsible for a
sharp decline in short-term visitors following the September 11 attacks. Several foreign
governments likewise complained that visa requirements and other security measures were
making it difficult for their citizens to travel to the United States. Using an empirical model
which distinguishes the impact of visa policy from economic and country-specific factors,
we find that changes in visa policy in the aftermath of 9/11 were not important contributors
to the decrease in travel to the United States. Rather, the reduction in entries was largest
among travelers who were not required to obtain a visa.
The views and opinions expressed in this paper are those of the authors and do not necessarily
represent official Treasury positions or policy. We thank Effi Benmelech, Scott Carrell, Karen
Dynan, Richard Falkenrath, Judith Hellerstein, Larry Katz, Ilyana Kuziemko, David Meyer, Harvey
Rosen, and seminar participants at the NBER Working Group on the Economics of National
Security for helpful discussions and comments. Cindy Soo provided excellent research assistance.
1 Introduction
The number of business and leisure travelers arriving in the United States dropped sharply
following the terror attacks of September 11, 2001. The number of non-immigrant visitors fell
by more than 17 percent for the period October 2002 to September 2003 (the government’s fiscal
year 2003) compared to the number of visitors in fiscal year 2000, and travel to the United States
had not recovered by late 2005 (Figure 1). In contrast, the number of legal permanent
immigrants did not change appreciably, though illegal immigration is believed to have declined
after 9/11 in response to heightened security measures.
1
This paper examines whether changes
in visa policy, which applied only to visitors from certain countries, were the key contributors to
the decline in short-term travel to the United States, or whether economic, psychological, or
other factors such as more stringent airport security had a larger impact on travel. Our results do
not apply to foreigners seeking to work or study in the United States, because the data do not
include longer-term arrivals such as students.
In the wake of the attacks, the U.S. government enhanced a wide range of border security
policies. Steps taken included both visible changes in security procedures at airports and other
entry points that affect all visitors, as well as changes in policies governing visa issuance that
affect only travelers who require a visa to enter the United States. The changes in visa policy
were not surprising in the wake of the attacks, given that the 9/11 terrorists had entered the
United States with legitimate visas.
Commentators in both the public and private sectors have claimed that tougher visa policies
are an important factor behind the decline in travel to the United States, and express concern that
post-9/11 visa policies hurt businesses by straining relationships with customers and hindering
opportunities for new business. For example, a recent Bloomberg article on the decline in
2
business travel to the United States after 9/11referenced an estimate by the National Foreign
Trade Council that tougher entry rules cost U.S. businesses $31 billion from 2002 to 2004 (the
article was also published by the International Herald Tribune under the title “’Fortress
America’ visa system scaring businesses away.”)
2
In late 2006, the London Times estimated the
cost of the decline in foreign tourists coming to the United States since 9/11 at $286 billion.
3
The article ties a large share of this cost to visa policy when it states, “The question is why
America is missing out. The immediate obstacle is the stricter security introduced since
September 2001.”
Suggestions that changes in visa policy played a fundamental role in the immediate post-
9/11 decline in foreign visitors also can be found on the web sites and in the press releases of
organizations ranging from think tanks to lobbyists. In December 2006, Tom Ridge, the first
Secretary of Homeland Security, was involved with a report from a travel industry group that
suggests, among other things, that the United States reassess the current visa and border control
regime.
4
The issue has also garnered attention from policymakers. A Federal Reserve Bank of
Dallas publication highlights companies’ concerns about the new post-9/11 visa requirements
(Orrenius 2003). The House of Representatives Government Reform Committee held hearings
on the topic in 2006, where even cello virtuoso Yo-Yo Ma chimed in, testifying that visa policies
were stifling cultural interchange. Foreign officials similarly have voiced complaints about post-
9/11 changes in visa policy.
1
Passel (2005) estimates that about 700,000 unauthorized immigrants entered the United States per year during
2000-2004, compared with 750,000 per year during 1995-1999.
2
Bliss and Hughes, “World’s ‘Worst’ Visa System Scares Business Away from U.S.”, December 25, 2006,
Bloomberg.com.
3
“Tourism Slump Worries U.S.”, The Sunday Times – Business, September 17, 2006.
4
Waterman, “Ridge Proposes U.S. Visa Reforms”, United Press International, December 22, 2006.
3
This paper assesses the impact of the post-9/11 visa regime on non-immigrant entries to the
United States. Our motivation is as follows. The changes in the visa regime were undertaken for
the purposes of national security. We take this benefit as given and will not seek to measure it.
As detailed above, however, many people weigh this potential benefit against the perceived cost
that changes in visa policy played an important role in the decline in temporary visitors to the
United States. Our goal is to introduce empirical evidence suggesting the scale of this cost.
We are aware of no recent economic analysis in the research literature that examines the
impact of economic, geopolitical, and security-related factors on business and leisure travel.
5
To
distinguish the effects of visa policy from other factors such as more general security measures
and the psychological impact of the attacks, we compare the impact of the 9/11 attacks on travel
by visitors that require a visa to enter the United States with the impact on visitors who do not.
To do this, we group countries by their participation in the visa waiver program, a section of the
U.S. legal code under which citizens of 27 countries (as of early 2007) are allowed to visit the
United States temporarily without first obtaining a visa. The countries included in the visa
waiver program are those whose citizens are deemed unlikely to pose a security threat and are
expected to leave the United States in line with immigration rules (Table 1). In 2003, entries
under this program represented roughly half of all overseas visitors to the United States (Siskin,
2004).
All travelers are subject to routine security restrictions such as examination of their
passports and luggage upon arrival into the United States, but only visitors from countries that do
not participate in the visa waiver program are affected by changes in non-immigrant visa policy.
Our econometric approach is based on this distinction: some changes in overall security policy
after 9/11 applied to all visitors to the United States, but changes in visa policy applied only to
4
certain travelers. The difference in treatment between travelers who must obtain a visa and those
who can enter without one provides a policy-induced variation by which to assess the impact of
changes in the visa regime that took place in the wake of the 9/11 attacks.
The analysis suggests that stricter visa policy did not play a direct role in reducing travel to
the United States in the two years following the September 11 attacks. After taking into account
economic and geopolitical effects, entries from countries requiring a visa did not fall by more
after 9/11 than visits by people not needing a visa. In fact, in the vast majority of our
specifications, the results indicate just the opposite – the decline was largest for travelers from
visa waiver countries whose citizens do not require a visa to enter the United States.
6
Several factors could explain the large decline in travel by visitors who did not require a
visa. One possibility is that the 9/11 attacks had a greater psychological effect on citizens of
countries in the visa waiver program, leaving them even more reluctant to travel to the United
States than citizens of other countries. Another possibility is that the heightened security
screening and other burdensome informal barriers that contribute to the “hassle factor” of travel
represented a proportionately larger incremental aggravation for travelers who did not require a
visa, resulting in a larger reduction in entries from these countries. Under this view, people
asserting that travelers requiring a visa were materially impacted by new obstacles after 9/11
might have been correct, but what they could not see was that travelers who did not require a
visa, such as nationals of France or the UK, were now being scrutinized as well. Since British
and French nationals received little scrutiny and suffered little aggravation before 9/11, the
5
There is a large literature on the economic influences of permanent immigration patterns. We also note that
Orrenius (2003) analyzes related issues for a general audience outside of a research framework.
6
We emphasize that our conclusion that visa policy was not the primary cause of reduced non-immigrant entries
does not apply to foreigners coming to the United States on worker or student visas. The factors behind a decision
to work or study abroad are likely to differ markedly from influences on business and leisure travel decisions.
Moreover, these visitors are not short-term and hence are generally excluded from our data. We thus provide no
evidence on this issue in either direction.
5
relatively greater change in treatment might thus be connected to the larger impact in their travel
to the United States.
In sum, after taking into account economic and country-specific factors, the post-9/11 drop
in entries was at least as large among travelers who did not need a visa, and in many
specifications, significantly larger. This indicates that changes in the formal visa process do not
explain very much of the decline in travel to the United States. What mattered instead were
either changes in foreign attitudes or changes in the post-9/11 security regime that affected the
decisions of all tourists and businesspersons considering travel to the United States, not just
those required to obtain visas.
2 Border Security Policy
Airport security changed substantially following 9/11, with passengers arriving in the
United States facing much greater scrutiny than before the attacks. Airlines are required to send
passenger lists in advance, exclusion lists are more vigorously maintained, and non-citizens are
required to provide digital fingerprints and have their photograph taken on arrival. Visa
approval takes longer.
State Department spokespersons have stated, however, that refusal rates did not change
significantly in the wake of the attacks. Indeed, though the refusal rate reached 35.1 percent in
fiscal year 2002 (following the attacks), the 31.7 percent rejection rate in fiscal year 2003 was
actually below the rate during fiscal 2001, the year ending in September 2001 (Clemens 2004).
7
These slight variations in visa rejection rates explain only a small portion of the 17 percent drop
in the total number of entries: the modestly higher rejection rate meant that several hundred
thousand more visa applications were refused in the year after the attack than in the year before
6
it, but the total number of entries fell by several million. Of course, it could be the case that
potential visa applicants, aware of the increased scrutiny they receive in the post-9/11 world,
self-select in a way that reduces the number of rejections made at U.S. diplomatic missions.
Independent from the criteria for granting or rejecting a visa application, potential travelers
requiring a visa are also affected by changes in consular procedures and higher fees that make
the visa process more arduous and expensive. Further, other security procedures affect all
visitors, regardless of whether or not they need a visa, such as the heightened airport security and
more involved screening for U.S.-bound flights that make travel less pleasant. Raising the cost
of boarding a plane and obtaining a visa would be expected, on the margin, to discourage
foreigners from undertaking business or leisure travel to the United States.
While some of the changes in visa policy are easily identifiable and were implemented on a
specific date, others have not been publicly disclosed or were implemented over time, including
being applied differently in different regions. Many of the changes, however, were instituted
shortly after the terrorist attacks. A State Department report notes that "the post-September 11,
2001 era witnessed immediate efforts to effect dramatic changes in CA's [Consular Affairs]
direction of the visa process” (Office of the Inspector General, Department of State 2004).
The most obvious change is that foreigners now pay more to apply for a visa to visit the
United States. Visa fees rose from $45 before September 11 to $65 in June 2002 and to $100 in
November 2002 (Rose 2004). Further, the time it takes to get approved for a visa increased for
many applicants. Shortly after the attacks, the Department of Justice requested that two new
name check procedures be added to the visa application process, requiring 20 and 30 days each.
The 30-day check under the so-called “Visa Condor” program applies to visa applicants from a
list of countries that is classified for national security reasons (GAO 2002). In early 2002,
7
These rates exclude some small visa categories, but the pattern remains essentially identical if these are added,
7
American consulates began to collect a supplemental application form from male visa applicants
aged 16 to 45 from every country. Visa applicants from certain countries must now undergo an
in-person interview at a U.S. embassy or consulate to obtain a visa.
These changes in visa policy, however, do not impact visitors from all foreign countries
because some visitors do not require a visa in the first place. The Immigration Reform and
Control Act of 1986 created the Visa Waiver Program to facilitate the entry of temporary visitors
from countries whose citizens were perceived as particularly unlikely to threaten U.S. national
security. Under the program, foreign nationals from participating countries (listed in Table 1)
are able to enter the United States for up to 90 days without obtaining a visa (nations
participating in the visa waiver program must extend reciprocal treatment to Americans).
Visitors are still checked against an exclusion list and must provide proof of a return ticket out of
the United States and adequate financial resources for their stay. A small number of visitors
from visa-waiver countries are required to obtain non-immigrant visas (for example, travelers
wishing to stay in the United States on a temporary basis for more than 90 days, visitors with
criminal backgrounds, and anyone who was previously denied a visa), and occasionally
foreigners unaware of their eligibility for the visa waiver program will unnecessarily apply for a
visa. These travelers, however, account for less than 10 percent of the non-immigrant entries
from most visa waiver countries in recent years. Based on this, we assume in the regression
analysis that all visitors from visa-waiver countries are without a visa, while all entrants who are
citizens of countries not included in the visa waiver program require one.
Countries were added to the visa waiver program starting with the United Kingdom in 1986
and most recently with Portugal, Singapore, and Uruguay in 1999. The program was made
rising through FY2002, but then falling in FY2003 to below pre-9/11 levels.
8
permanent in 2000 and at that time included 29 countries. Argentina and Uruguay were removed
from the program in 2002 and 2003, respectively.
The visa waiver program has come under review in the wake of the 9/11 attacks and
subsequent incidents, as concerns have arisen that terrorists or other criminals could exploit the
program. After all, Zacarias Moussaoui, a French national convicted as a co-conspirator in the
September 11 attacks and Richard Reid, a British citizen convicted of trying to detonate a bomb
concealed in his shoes while on a flight to the United States, both entered the United States under
the Visa Waiver Program. To date, however, the program has not been changed, though efforts
to improve the list of foreign nationals ineligible for entry to the United States might be seen as
an attempt to address the potential vulnerability.
3 Influences on Business and Leisure Travel
A change in visa policy was only one of several factors affecting foreign travel to the
United States. The September 11 attacks had a significant impact in reducing airplane travel to
the United States that was independent of economic factors and security measures; one might
term this a “psychological” impact in making people reluctant to fly in the immediate aftermath
of the attacks. This is mirrored by the reduction in domestic U.S. air travel after 9/11: according
to Department of Transportation data, the number of passengers on domestic non-stop segments
in the United States was 11 percent lower in fiscal 2002 than in fiscal 2000 (Figure 2).
Economic factors affect travel as well. Strong U.S. growth would likely increase business
opportunities that cause foreign businesspersons to travel to the United States, and the economic
strength or weakness in travelers' home countries might similarly affect business and leisure
travel. Exchange rate fluctuations change the effective cost, denominated in foreign
consumption units, for foreigners considering travel to the United States. Finally, the onset of
9
the war in Iraq appears to have led to a drop-off in travel in early 2003 as potential visitors
awaited the outcome of the conflict.
There has been limited research on the economic factors that influence business and leisure
travel. The most closely related studies examine the determinants of total spending by
international tourists, which equals the product of the number of visits (our focus) and the
average amount spent on each visit. Rhomberg and Boissonneault (1964) and Gray (1966)
estimate elasticities of demand for spending by international travelers with respect to national
and per capita income and exchange rates and find that the response of spending to these factors
is statistically significant. Gray also finds that the cost of transportation was not a significant
factor influencing international travel, though of course this and the other results could have
changed in the intervening four decades. Kwack (1971) likewise shows that spending abroad by
travelers responds substantially to changes in national incomes and real exchange rates and finds
that these elasticities are higher for foreign travelers entering the United States than for U.S.
travelers visiting foreign countries (though, again, these results apply to data from the 1960’s).
More recently, Di Matteo (1993) and Vilasuso and Menz (1998) find that national income and
the exchange rate are the key determinants of spending by Canadians traveling in the United
States.
An analytic framework to assess the influences on the number of business and pleasure
travelers would model the choice of individuals of whether or not to travel to the United States,
with this decision influenced by personal circumstances, including both financial and otherwise.
Since the requirement of individual-level data on potential travelers is unrealistic, we estimate
instead a macroeconomic framework in which the dependent variable is the monthly number of
entries to the United States from individual countries of origin. The explanatory variables are
economic factors such as overall GDP growth in the United States and each partner nation and
changes in the real exchange rate between the two countries. We also take into account
10
influences on travel patterns such as the season in the United States (the hemisphere of the origin
country turns out not to matter), the war in Iraq, and whether a traveler requires a visa to visit the
United States.
The econometric analysis is carried out on a panel of entries into the United States from
each of up to 65 countries, denoted as i. This is akin to using data on bilateral trade between the
United States and various partners. Country fixed effects are included to capture country-
specific factors, while monthly fixed effects capture common factors that change over time. The
specification includes a post-9/11 dummy variable, a dummy to indicate whether travelers from
each country are eligible for the visa waiver program (VWP), and the interaction between the
9/11 and the visa waiver indicators. This allows for a “differences-in-differences” framework to
assess whether post-9/11 changes affected entrants from VWP and non-VWP countries
differently: all travelers to the United States were affected by the post-9/11 environment and by
some changes in security procedures, but only travelers from non-VWP countries were affected
by changes in visa rules. The empirical specification allows us to assess whether changes in
travel patterns after 9/11 differed between travelers who needed a visa and those who did not.
4 Data and Specification
Monthly data on non-immigrant entries to the United States were obtained from the U.S.
Department of Homeland Security (DHS). The data include I-94 admissions only, meaning it
excludes the majority of short-term land-border admissions from Canada and Mexico. We
follow DHS in focusing on these data because the large number of Canadian and Mexican entries
would swamp non-immigration trends from the rest of the world that are more relevant to visa
policy. Further, the motivating factors for crossing a land border may differ from the factors
behind longer-distance travel of nationals beyond the immediate U.S. neighbors.
11
These unpublished data are compiled by DHS’s Office of Immigration Statistics and include
the monthly number of entries for fiscal years 1996 and 1998 to 2003 (as an example, fiscal year
2003 runs from October 2002 to September 2003).
8
There are some gaps in the data, with only
annual figures available for 1995 and no data at all for 1997. We have not been able to obtain
monthly data for years after 2003. The data include the number of entries by class of admission
(e.g. tourist or businessperson) and by country of citizenship.
9
This last classification means
that, for example, a French national arriving in the United States is counted as French regardless
of whether he or she resides in France and regardless of the port of embarkation.
The number of entries consistently trended upward from the early 1990s until September
2001. After the post-9/11 drop-off, travel to the United States slowly grew back to near its
former peak. Non-immigrant travel had not reached its pre-9/11 level by the end of September
2005, the most recent year for which we have data (that is, DHS has published the total number
of entries for fiscal years 2004 and 2005, but has not made available monthly data by country).
The panel unit root test of Im, Pesaran, and Shin (2003) rejects the null of non-stationarity for
both the levels and log of entries when all countries are pooled together for the years in which
continuous data are available from October 1997 to September 2003 (the null is rejected with
and without including a trend in the test).
We drop countries lacking economic data such as measures of output and inflation on at
least a quarterly frequency. The resulting sample includes 65 countries and covers over 86
percent of all entries in fiscal year 2003. Our sample contains 23 countries that were participants
at some point in the visa waiver program – a few countries move in or out of the visa waiver
8
Most of these data have been aggregated into annual statistics and included in the Office of Immigration Statistics’
Yearbook of Immigration Statistics, but the monthly statistics on country-by-country entries are not published
separately (the unpublished data we use are from DHS Table 607).
9
The split between business and pleasure is not always available and in some cases is imputed, preventing us from
further including these characteristics in our analysis.
12
program during the years in our sample.
10
We have tested our results with a constant panel,
omitting these countries as well as others that enter the data set after 1995. Though this
procedure requires omitting a large number of observations, it has little qualitative impact on our
results.
We further exclude three countries from our primary specifications, though we report full
results with their inclusion in the appendix. First, we exclude Argentina because the very close
timing of its economic and political crisis, its removal from the visa waiver program, and the
9/11 attacks potentially poses problems for our identification strategy. We also exclude
Canadian and Mexican entrants because our I-94 data excludes land-border crossings and hence
captures only the small share of entrants from those nations who arrive from another country or
via air travel. Further, the number of Mexican visitors requiring visas is extremely large relative
to the number of other entries that require visas, so it has the potential to swamp developments in
other countries. In Appendix Table A1, we provide results for the cases when all countries are
included, when only Argentina is removed, and when only Canada and Mexico are removed.
Our basic results holds for all of these cases. When including Canada, we consider it along with
the visa waiver countries, even though this is not formally the case, because nearly all Canadians
enter the United States without a visa under a different agreement than the visa waiver program.
The data indicate that both international travel to the United States and domestic travel
within the United States declined considerably following the 9/11 attacks (Figures 1 and 2).
11
The recovery of domestic travel, however, was far more rapid and steady, suggesting the
possibility of different influences on international travelers. The total number of entries in the
10
Argentina was removed, while Australia, Slovenia, Portugal, and Singapore were added.
11
There is a very strong cyclical component to monthly travel data. Graphs such as Figure 1, Panel B, that are
labeled “seasonally adjusted” have had monthly variation stripped out by adjusting the raw series by the coefficients
from regressions of the log series on month dummies. The short time series involved, however, mean that this
13
aggregate data from all 204 countries declined by more than 17 percent from the last full fiscal
year before the attacks to fiscal 2003 (Figure 1). Table 2 shows that the corresponding drop in
entries from our sample of 65 countries is also 17 percent. This includes a 22 percent decline in
entries over this period from the 23 visa waiver countries in our sample, and an 11 percent drop
in entries from non-visa waiver countries.
12
Our sample of countries includes a very high percentage of entries from visa waiver
countries, which implies that the 22 percent drop in entries from these countries in our data is
close to the decline in the aggregate data for all visa waiver countries (that is, including countries
not in our sample). Coupled with the facts that aggregate entries dropped 17 percent and that
entries from visa waiver countries accounted for about 55 percent of all entries over this period,
this implies the aggregate drop from all non-visa waiver countries was about 11 percent—the
same as in our sample. This suggests that our dataset is representative of entries from both visa
waiver and non-visa waiver countries.
The summary statistics in Table 2, both for the aggregate data as well as for the countries in
our dataset, do not indicate that visa policies contributed substantially to the drop in entries;
indeed, the raw numbers suggest the opposite in that the average decline was larger for entries
from countries whose citizens did not require a visa. Figure 3 plots seasonally adjusted entry
data from a constant panel of all countries whose visa-waiver program status did not change
from 1996 on (including those for which we have no economic data). The figure indicates that
both the trend growth before 9/11 and the subsequent response to the attacks is remarkably
similar from citizens of the two groups of countries. If anything, travel from countries not
procedure is highly imperfect. Small or short term fluctuations or correlations in fluctuations should not be taken too
seriously.
12
As discussed above, we exclude Argentina, Canada, and Mexico from our baseline regressions, and so have also
excluded them in these calculations. Given Mexico’s small increase in travelers, including these countries would
make the relative drop in visa waiver entries even larger.
14
requiring visas exhibit the more significant negative response. Hence, just from looking at these
data, one might conclude that the reduction in business and pleasure travel resulted from the
psychological or economic effects of the terrorist attacks that were common to all countries, or
from general security measures impacting all travelers rather than from changes in visa policy.
While the summary statistics suggest that the impact of visa policies on non-immigrant entries
was limited, a multivariate regression framework is required to distinguish the influence of visa
policies from economic and other factors impacting travel decisions.
Our base specification for entries from country i in month t takes the form:
12 3
ln( ) ln ( * -9/11)
ln
it i t it it
it
VWP
Entries Country Time GDP VWP Post
RER
β
βψ β
⎡⎤
⎢⎥
=++∆+
⎢⎥
⎢⎥
∆
⎣⎦
ε
+
(1)
where we include country and time dummy variables. Data on real GDP, inflation, and nominal
exchange rates in terms of local currency per dollar are from the IMF's International Financial
Statistics (IFS) database. The economic and indicator variables are at a monthly frequency; for
real GDP, we linearly interpolated quarterly data to obtain monthly values.
13
The real exchange
rate, RER, is calculated using the nominal exchange rate (in foreign currency per dollar) and
U.S. and foreign consumer price inflation; a larger value thus represents a real appreciation of
the U.S. dollar. Real GDP growth and the real exchange rate are included as 12-month growth
rates. The VWP variable is one for countries participating in the visa waiver program and zero
otherwise. Since the specification includes country fixed effects, the VWP coefficient by itself is
estimated entirely from time series variation in countries that transition in or out of the program.
The specification uses the log of the number of entries along with country fixed effects, so it
matches changes in entrants in percentage terms against changes in economic variables, also in
15
percentage terms. As such, the coefficients on economic variables provide easily interpreted
elasticities. Time dummies capture changes in aggregate entries from one period to the next.
Estimating with the log of entries is somewhat problematic, however, in that it gives the
same weight in the regression to a 10 percent change in visits from a country with many travelers
such as Japan as it does a 10 percent change in visits from a country with relatively few entries
such as Kazakhstan. Since our goal is to understand the relationship between changes in visa
policy and the number of non-immigrant entrants to the United States, in our baseline
specification we weight the regressions by the level of entrants for each country in fiscal 1996.
14
This approach makes sense when one is concerned about heterogeneity in the true underlying
relationships across countries and wants to put more weight on the estimated relationships for the
larger countries (that is, to give more weight to the GDP elasticity of visits for Japan than for
Kazakhstan).
It is also interesting, however, to consider the case without weights, which makes sense if
one believes the source of error in the estimates is a country-month specific unobservable
shocks; the unweighted regression treats each observation as yielding the same amount of
information about the impact on visa policies. We see the weighted regressions as more relevant
to this analysis, but report both weighted and unweighted results for all specifications. Finally,
in Appendix Table A2, we also consider alternative weighting schemes -- population, PPP-
adjusted per capita GDP levels, and bilateral trade with the U.S. in 1995. These weighting
schemes are broadly consistent with our results in that none of the interaction coefficients are
positive and significant at 5% or better. We prefer the baseline weights, however, because each
13
The consumer price index for Australia is available only by quarters in some years and is linearly interpolated to
obtain monthly values.
14
Using weights that are in any way related to the independent variable can be problematic, particularly if there is a
time-series element to the weights that allows them to be correlated to the error term in the regressions. Here, we
16
of these alternatives greatly skews countries relative to their import to U.S. tourism. For
instance, India and Singapore have disproportionately large impacts on the population and per-
capita GDP weighted regressions, respectively.
In all of the regressions, we assume that the causality runs from macroeconomic variables
of GDP growth and the exchange rate to the number of entries, and that the number of visits
from any one country is not large enough to affect aggregate growth rates or currency markets.
This is based on the observation that even the high water mark of roughly 3.5 million temporary
visits from all countries in July 2000 is modest compared to the U.S. population or labor force,
and that the spending of these visitors and thus their impact on the exchange value of the dollar
is dwarfed by the trillions of dollars in daily turnover in foreign exchange markets.
The coefficient on the interaction of the post-9/11 and VWP indicators measures the
differential impact of 9/11 on people from visa waiver countries compared to visitors who must
obtain a visa. A significant positive value on this interaction term would indicate that,
conditional on the economic factors and other controls, post-9/11 changes were associated with a
larger reduction in entries from non-visa waiver countries than visa waiver countries. Indeed,
proponents of the “tough visa hypothesis” would predict a quantitatively large and statistically
significant coefficient on this interaction, suggesting stricter visa policies applying only to non-
VWP entries greatly affected the level of entries. In fact, the vast majority of our results
generate a negative coefficient on this interaction term, suggesting the decline in travel to the
United States after 9/11 was at least as pronounced in countries whose citizens did not require a
visa. This negative coefficient is statistically significant in many specifications, including both
weighted and unweighted regressions.
worry less about this because our data is in percent changes at a monthly frequency, and hence the error term is
unlikely to be correlated to the weights, which are the levels of entrants in only the first year of our dataset.
17
To be sure, we do not interpret this negative coefficient to mean that stricter visa policy led
to increased entries from affected countries—it is not reasonable to believe that the demand for
travel increases when it becomes more costly. Rather, we are suggesting that the factors behind
the decline in visits to the United States were either unrelated or negatively correlated to the
tightening in visa policies after September 11.
5 Regression Results
The first half of Table 3 provides results from our baseline specification panel regression (1)
on the influences of the number of entries to the United States. Column (1) shows the baseline
specification in which we weight by a previous annual level of entries, and column (2) shows the
identical specification with no weighting. In addition, we include results for specifications that
include country-specific linear trends in columns (3) and (4). We consider these specifications
less meaningful, since the overall pattern of entries shown in Figure 3 does not suggest important
differences between the groups in terms of long-term trends and the limited time series is not
sufficient to precisely estimate a stable differential trend between the two groups of countries.
Further, since countries appear in our dataset over different periods of time and we are entirely
missing the data for fiscal 1997, these trends may inaccurately impose unrealistically large
differences in country growth rates. Standard errors in all specifications are robust to
heteroscedasticity and are clustered by country to control for serial correlation within each
country.
15
15
Bertrand, Duflo, and Mullainathan (2004) discusses how positive serial correlation in outcome, treatment, and
control variables can lead to underestimated standard errors on diff-in-diff interaction coefficients. They show that
clustering at the group (country in our case, state in theirs) level can alleviate the problem to some extent where
there are a large number of groups (such as 50 states, or in our case, 62 countries). Further, while the possibility
that our standard errors are underestimated makes it more likely we would find a significant and negative
coefficient, it also makes it more likely that we would find a “false” positive interaction coefficient. This makes our
result that there are essentially no positive and significant coefficients in our specifications even more striking.
18
None of these baseline four specifications show a positive and statistically significant
interaction term (the “difference-in-differences” coefficient), as would be expected if visa
policies were a particularly important contributor to the drop in entries. This is because a
positive interaction coefficient would mean that relative to levels before 9/11, countries not
requiring visas generally had a larger number of travelers compared to countries that do need
visas. If anything, however, the results suggest the opposite. The interaction coefficient is
negative in three of the four specifications and significant at the 5 percent level in two of the
three. For example, the interaction coefficient in column (1) suggests that, after taking into
account country-specific economic variables and other common influences, countries in the visa
waiver program typically had a 19 percentage point larger decline in visitors to the United States
than countries for which visa policy would be expected to matter. The lone positive interaction
coefficient, in column (4), is small quantitatively and highly insignificant. These results imply
that changed visa policies were not a particularly important driver of reduced entries, or even
that unobserved factors contributing to the drop in travel to the United States are negatively
correlated with changes in visa policies. This could be the case, for example, if visa policies
were tightened in countries least affected by the increased “fear of flying” after 9/11 that perhaps
had a larger impact on potential travelers from advanced economies who did not need a visa to
enter the United States.
Other coefficients are consistent with the existing literature on the determinants of travel
spending. Stronger growth in each home country is associated with increased travel to the
United States, and the elasticity is quite large. The coefficients on the real exchange rate are
uniformly negative as expected, implying that a stronger dollar (a real depreciation of the
currency of the origin country) leads to less travel to the United States. This is in line with the
expectation that a weaker currency in the origin country makes travel to the United States (an
import of U.S. services) and the dollar-denominated fee to apply for a visa more expensive for
19
foreigners. The precision of these estimates is substantial, with only one insignificant coefficient
across these eight point estimates.
The visa waiver dummy is positive in most specifications, as expected, but is not
statistically significant. This is not particularly surprising given it is identified entirely from time
series variation in the small number of countries that enter the visa waiver program in the middle
of our dataset.
Our results go against the conventional wisdom that post-9/11 changes in visa policy had an
important impact on travel to the United States. It should be kept in mind, however that our
empirical approach is affected by a number of statistical issues that affect difference-in-
differences regression specifications. An ideal setup for such analyses would involve a truly
random selection of the “treatment” group—in our case, the group of countries not in the visa
waiver program whose travelers were thus affected by changes in visa rules. It is clearly not the
case, however, that countries in or out of the visa waiver program are randomly selected. As
such, we next add additional conditioning variables as well as run regressions on subsets of the
countries to determine the robustness of the lack of a positive and significant interaction
coefficient.
We start by adding the real growth in expenditures on travel and tourism by citizens in each
of the foreign countries as a new covariate. These data are compiled annually by the World
Travel and Tourism Council along with the consulting firm Accenture and include all
expenditures by domestic business and pleasure travelers on domestic and foreign travel.
16
We
linearly interpolate these annual growth rates and add them to condition on country-specific
shocks in demand for any form of travel, not just travel to the United States.
17
As these are
16
We downloaded these data from www.wttc.org.
17
We are implicitly assuming that the bulk of these travel expenses in all foreign countries is accounted for by
travel to domestic and non-U.S. foreign destinations.
20
growth rates, there are a few substantial outliers, so we drop any country-year observations with
a magnitude of change exceeding 20 percent up or down—this screens out about 5 percent of the
observations.
Table 4 columns (1) to (4) show results from our original four specifications, modified to
include this proxy variable for the overall demand for travel. Though this variable is
extrapolated from annual data, while our growth series is extrapolated from quarterly data, one
may reasonably be concerned that simultaneous inclusion of both variables gives rise to
endogeneity problems. As such, columns (5) to (8) give the results from regressions that include
this new travel demand variable but not home country growth. As expected, the variable
generally enters with a positive, though typically insignificant, coefficient. Most importantly for
our results, however, the interaction coefficients in Table 4 remains negative in most of the
specifications (significantly so, in several columns, for at least the 10% level), and is never
significant and positive .
Next, we trim the dataset in order to create a more comparable set of countries. For
instance, the poorest country in the dataset would be a highly unlikely addition to the visa waiver
program in the same way that it would be unusual for the richest country to be excluded. We
thus estimate specifications that include only those countries with purchasing-power-parity
(PPP) adjusted per capita GDP levels in 2000 ranging from 39 to 66 percent of the per capita
GDP of the United States. This includes all countries for which we have data that range in per
capita wealth from the Czech Republic to Finland, inclusive. This is among the only ranges in
the data where there is a comparable number of visa-waiver and non-visa waiver countries.
Table 5 lists the 12 countries falling in this range, organized by membership in the visa waiver
program, as well as the regression results for this restricted set. Once again, none of the point
estimates on the interaction terms are positive and significant. In fact, all are negative and two
are significant at the 10 percent level.
21
Next, we employ a more formal procedure to achieve the same goal of a more comparable
set of visa-waiver and non-visa waiver countries. We generate propensity scores for each
country that represent the likelihood that each country would be included in the visa-waiver
program based on fixed country characteristics, including the population in 1995, amount of
trade with the United States in 1995, PPP-adjusted GDP in 1995, the share of business travelers
in 2000, and the natural log of the physical distance from the United States.
18
We then omit
from our regressions all countries in the visa waiver program with propensity scores above that
of the country outside of the visa waiver program with the highest scores. Similarly, we exclude
countries requiring visas that have scores lower than the lowest visa waiver country. This
eliminates slightly less than half of the countries in the dataset.
Panel A of Table (6) reports the results from the initial propensity score probit regression.
As one might expect, it indicates that countries that are larger, have more trade with the United
States, higher per capita GDP, and that are closer to the United States are more likely to be
members of the visa waiver program. The probit coefficient on the share of business travelers is
insignificant. Panel B gives the list of countries with sufficiently close propensity scores to be
kept in the dataset as well as the results from the difference-in-differences regressions we run on
this filtered dataset. As was the case with all other specifications, there is again no regression in
which the interaction between the 9/11 and visa waiver indicators is positive and significant.
Three of the four coefficients are negative, though none of the four is statistically significant.
Having demonstrated the lack of positive and statistically significant impact of changes in
visa policy on travel to the United States, we next show in Figure 4 plots of the seasonally
adjusted and demeaned time fixed effects when we estimate (1) with no interaction term, and
allow for distinct time fixed effects for the visa waiver countries and the non-visa waiver
18
Log distance was used instead of raw distance in order to satisfy the balancing property in the propensity score
22
countries. These series were calculated as the residuals of regressions of the actual fixed effects
on month dummies and a constant term. Hence, the relative vertical levels of the series have no
information, but the time series variation indicates how the seasonally adjusted common
component of each group changes over time. As before, this method for seasonal adjustment on
a short time-series is somewhat crude and too much emphasis should not be placed on small
scale or high frequency movements. Aside from the general upward trends in the lines, one can
clearly see sharp drops in the wake of the 9/11 attacks and during the 3 months of major combat
operations in Iraq in early 2003.
Panel A of Figure 4 contains our baseline specification where observations from each
country are weighted by the level of entrants from that country in 1996. The plot shows that
both series had essentially the same long term trend, rising about 30 percent from early 1996 to
late 2000 and early 2001. During 2001, before the 9/11 attacks, both series declined somewhat,
with the decline in travel by citizens of visa waiver countries somewhat steeper than travel from
countries requiring visas. The real plunge in entries in both groups of countries, however, occurs
after the 9/11 attacks. Though the gap is not striking relative to differences at other time periods,
this plot makes it clear that the visa-waiver countries saw at least as large a negative impact on
short-term travel to the United States after 9/11 as the non-visa waiver countries. From levels
that look highly similar from early 2000 to late 2001, entries not requiring visas remain
persistently far below levels in countries that do need visas. This is what results in the negative
interaction term in our difference-in-differences estimates.
6 Caveats
regression.
23
While the previous section provides the most sensible specifications to test the impact of
visa policies on overall entries to the United States, several additional caveats to our results
should be kept in mind. We have focused on the question of whether visa policy played an
important role in the post-9/11 decline in entries, and not on the question of whether changes in
visa policy significantly reduced entries for any particular country or resulted in the granting of
fewer visas (independent of their use). Our dataset, which measures entrants, is most appropriate
for answering this question because it correctly puts more weight on multiple entrants using the
same visa, and compared to a dataset purely covering visa issuance, it offers a comparison to
similar changes in countries without visa requirements. It does not, however, allow us to study
the granting of visas directly (beyond noting that visa rejection rates have not changed
significantly).
19
Our focus on the question of assessing the impact of visa policy on total travel to the United
States leads us to emphasize the weighted regressions we have presented as a baseline,
attributing more attention to the larger number of visitors from, say, Spain and Chile, than the
much smaller number of travelers from, say, Luxembourg or Georgia. Interest in entries from
smaller countries, however, might call for greater attention to the unweighted regressions. Panel
B of Figure 4 plots the separate time fixed effects for this unweighted case. Differences between
the cases in Panel A and B necessarily reflect differences in entry patterns between countries
with many entries and countries with few. Indeed, though the two series move largely in parallel
from around 2000, the earlier differential trend in the unweighted case suggests that there is a
noticeable “visa effect” for some of the smaller countries. In essence, because some small
countries, predominantly in the non-visa waiver set, have faster entry growth rates (though the
19
Further, some visas have long lives and hence, some travelers that were previously approved under a cheap visa
regime can continue visiting the U.S. after 9/11 without undergoing the new visa procedures. We think this applies
to a small number of entrants.
24
differential trend seems to decline), one might have expected their fixed effect series to rise even
further above the visa waiver series in the counterfactual post-9/11 era. The gap, however,
remains fairly constant (and shrinks at the end), implying that visa policies may have restricted
growth in non-visa waiver program entries from these small countries—travel from these nations
was growing rapidly but this was throttled back after 9/11. This dynamic is surely the cause of
the positive (though still statistically insignificant) interaction coefficient for the unweighted
specification with separate country trends in Table 3.
20
Hence, while we find no meaningful
contribution of visa policies to the overall decline in travel to the United States, our analysis does
suggest that what would otherwise have been rapid growth in entries from a few small countries
was hampered by post-9/11 changes in visa policies.
A final caveat to our results is that we only have data for two years following the 9/11
attacks. The plots in both panels of Figure 4 suggest that the drop in travel by citizens of
countries in the visa waiver program abated somewhat toward the end of our dataset. When we
run our baseline regression but divide the post-9/11 variable into two separate variables for 2002
and 2003, we find an even more significant difference between the two groups of countries in
fiscal year 2002. This further strengthens our result that post-9/11 changes in visa policy do not
account for the decline in travel to the United States, since the sharpest decline in fiscal year
2002—just after the attacks—was by travel of people who did not need a visa. This difference is
substantially attenuated, though, by the second year after the attacks and its statistical
significance is often lost.
21
To be clear, this still means that by the end of fiscal 2003, there was
no support for the hypothesis that tighter visa policy reduced travel. If the trend from 2002 to
20
A specification with two different quadratic trends for all of the visa waiver and all of the non-visa waiver
countries also results in a positive, though small and statistically insignificant, interaction term. Given the
differences between countries even within the two groups, it is more sensible to estimate country-specific rather
than group-specific time trends.
21
These results are available from the authors upon request.
25
2003 were to continue, however, it is plausible that in later years one might detect an impact of
these visa policies on aggregate entries to the United States. Data for arrivals in fiscal years 2004
and beyond would be needed to examine this possibility.
7 Other Factors Influencing Post-9/11 Travel
We have shown that for a wide variety of specifications, and with only a few minor caveats,
there is no evidence that entries to the United States of travelers requiring visas dropped by more
after 9/11 than travel by people not requiring visas. The evidence suggests that, if there is any
difference between the groups, it is that the number of entrants from countries outside of the visa
waiver program dropped by less. Since it is implausible that tighter visa policies led to more
visitors, we next consider other variables, unrelated to visa policy, that might help explain some
of the cross-country variation in the response of potential travelers after the 9/11 attacks.
We start by considering heterogeneity across countries in terms of the composition of
travelers between businesspeople and tourists. Our dataset includes an imputed estimate for each
country of the number of business travelers as a percent of total non-immigrant entries in 2000.
We assume that this composition is relatively stable over time and so include it as an interaction
term with the 9/11 indicator that equals zero prior to 9/11 and one after the attacks. In columns
(1) to (4) of Table 7, we show the regression results when this business traveler interaction term
is included. Since the demand for business travel is likely to be less elastic than pleasure travel,
the positive coefficient on this interaction term is unsurprising. As travel became more
expensive for everyone after 9/11, business travelers were less likely to respond and change their
behavior than tourists were, so that business travel overall declined by less than non-business
travel. This estimate is insignificant for the baseline case without country-specific time trends,
and does not change the result on the visa waiver interaction term. It does have significant
explanatory power for the case with time trends.
26
Next we consider the impact of different attitudes toward the United States in accounting
for changes in the demand for travel. Ideally, one would like to have a time series show attitudes
before and after 9/11. We do not have such data, but instead must use a snapshot of foreign
views of the United States taken in early 2002 by the Pew Global Attitudes Project. This project
asked citizens of 44 countries about their view of the United States (among many other things).
22
We assume that the 9/11 attacks did not change the cross-country ordering of views toward the
United States but rather increased the dispersion in such views, so a snapshot of post-9/11 views
would be highly correlated with the change in views from an earlier period. The average survey
response for each country on this question is a number between 1.0 and 4.0, with a score of 3.9,
for instance, representing a case where most recipients held very unfavorable views of the
United States. Unfortunately, the survey data cover only 17 of the countries in our dataset.
Nonetheless, as shown in columns (5) through (8) of Table 7, this interaction coefficient enters
negatively and with a high statistical significance in most of our specifications. The negative
coefficient implies that travelers from countries with a less favorable view of the United States
entered with a lower frequency following the 9/11 attacks compared with before the 9/11 attacks.
This suggests that differing perceptions mattered for travelers’ decisions. With only 17 countries
included in the regression, however, we cannot provide more definitive evidence. Further,
inclusion of this variable pushes the 9/11 and visa waiver program interaction coefficient even
further negative, and hence, itself cannot explain the lack of observable “visa effect.”
Following on this result, and in light of many articles suggesting that the view of the United
States after 9/11 has worsened in countries with large Muslim communities, we add a variable
that captures the percent of the population that is Muslim in each of the countries in our dataset.
The data are taken from the CIA factbook, available online, which generally provides a specific
22
Specifically, question 61 asks, “Please tell me if you have a very favorable, somewhat favorable, somewhat
27
percentage for the Muslim population (different types of Muslims, as well as any mention of
Islam, were added together for this measure). When it does not, the CIA factbook generally lists
several religions as occupying a subset of the population. In these cases, we divide up the
residual percentages among the religions mentioned, with slightly higher percentages going to
groups listed earlier. For instance, if an entry said “65% Christian, 25% Hindu, others Muslim
and Jewish”, we allocate the unattributed 10 percent by coding 6 percent as Muslim and 4
percent as Jewish (because Muslim was listed first). In all cases with no mention of Muslims or
Islam, we assume their population share is zero.
Columns (1) to (4) of Table 8 show that for every specification, the term capturing the share
of Muslim populations interacted with post-9/11 is negative and highly significant. Here it is
worth noting that this interaction cannot be a particularly significant quantitative driver of the
post-9/11 drop in entries since most of the large countries by entries have small Muslim
populations. This coefficient does capture, however, that the few large Muslim countries
generally saw relatively large declines in entries, even when compared to similar countries that
also required visas. We formalize this by including a triple interaction term in columns (5) to
(8), which captures variation by country in or out of the visa-waiver program, after 9/11, by the
size of their Muslim population. We do not find any significant results for the triple interaction
term. There are few countries in the visa waiver program with substantial Muslim populations,
so it is not surprising that we do not find a significant effect. The large unconditional declines
from a country like Singapore, though, without the need for visas and with a large Muslim
population, suggests that the tendency for countries with large Muslim populations to have less
entries after 9/11 is independent of visa policy. As with the “view of U.S.” variable, however,
unfavorable or very unfavorable opinion of The United States.”
28
accounting for variation in Muslim populations in the data also fails to reverse the sign on the
9/11 and visa waiver program interaction term.
Further discussion of these relationships is outside the scope of our analysis, but we
conclude that factors far broader than visa policy, such as whether a traveler is a business or
pleasure traveler, or their views of the United States, are more likely to explain the differences
across countries in the drop in travel to the United States after 9/11 than changes in visa policy.
8 Conclusion
We find that the decline in visits from countries requiring a visa was not larger than the
decline in entries from countries exempt from visa requirements. This suggests that tighter visa
policy was not the cause of the sharp drop in business and leisure travel to the United States in
the wake of the 9/11 attacks. Changes in the visa process enacted with a delay after 9/11 could
be connected with the complaints voiced about visa problems, and these changes perhaps
reduced entries somewhat from countries outside the visa waiver program. The timing of this
impact, however, does not correspond with the immediate drop in entries, nor does its magnitude
stand out when compared with the impact of changes unrelated to visa policy.
We emphasize that our results, taken on their own, do not suggest the correct set of visa
policies. Policymakers must jointly weigh the costs and benefits of these policies, while our
paper sheds light only on the cost side of this equation. Further, our data only allows us to
measure this cost as a function of the number of foreign entries. Hence, our analysis excludes
any costs generated by changed visa policies that do not materialize in a forgone visit, such as ill
will toward the United States. Several of the groups that have called attention to what they see
as a relationship between stricter visa policies and reduced entries offer solutions such as
“greeting international travelers with a smile” at airports or improving the efficiency at passport
29
checkpoints in international terminals. Our results do not cast any doubts on efforts such as
these, which apply to all entrants regardless of their country of original.
It is difficult to say for sure why the travel plans of people who did not require a visa to visit
the United States were affected more (or, at least, not less) after 9/11 than travel by people who
needed to obtain a visa. In addition to some of the factors considered in section 7, another
explanation for the larger drop in visitors who do not require visas might be connected to the
increase in non-monetary costs, such as waiting times and other aggravations associated with
increased security involved in traveling to the United States. Before 9/11, short-term visits to the
United States were nearly hassle-free for citizens of countries participating in the visa waiver
program. In contrast, nationals of other countries, mainly less-developed ones, faced the hazards
of the visa application process. After 9/11, visa applicants might well have received greater
scrutiny, and indeed, many had to wait longer for visas and travel further for an in-person
interview at a U.S. diplomatic post. Travelers who did not require a visa, however, also faced
new costs—new hassles, one might say—to enter as a result of post-9/11 changes in security.
Compared to the previous near-zero amount of hassle, the added aggravation for travelers from
these visa waiver countries was proportionately enormous. While changes in visa policy might
have affected travelers needing them, this appears to have been a secondary factor in accounting
for the overall decline in short-term business and tourist travel to the United States.
Future research could shed light on the causes of the steep reduction in temporary entries of
businesspersons and tourists to the United States after the 9/11 attacks. The results in this paper,
however, lead us to dismiss the idea that changes in visa policy are primarily to blame.
30
References
Clemons, Steven C. (2004). “Land of the Free?” Op-Ed in The New York Times, March 31.
("United States Visa Application Rejection Rates" found at
http://www.steveclemons.com/visafees.htm, June 15, 2004.)
Bureau of Consular Affairs, Department of State (2004). "Initiatives by the Bureau of Consular
Affairs to Enhance National Security -- Fact Sheet, September 5, 2002" found at
http://www.state.gov/coalition/cr/fs/13316.htm, May 3, 2004.
Office of Immigration Statistics, Department of Homeland Security (2004, 2005). Yearbook of
Immigration Statistics.
Office of the Inspector General, Department of State (2004). "Review of Nonimmigrant Visa
Issuance Policy and Procedures (ISP-I-03-26)" found at
http://oig.state.gov/oig/lbry/isprpts/domestic/26634.htm, May 3, 2004.
Di Matteo, Livio. (1993). "Determinants of cross-border trips and spending by Canadians in the
United States." Canadian Business Economics.
Bertrand, Duflo, Mullainaithan (2004). “How much should we trust Differences-in-Differences
Estimates?” Quarterly Journal of Economics.
General Accounting Office, United States (2002). "Visa Process Should be Strengthened as an
Antiterrorism Tool." October.
General Accounting Office, United States (2002). "Implications of Eliminating the Visa Waiver
Program." GAO-03-38. November.
Gray, Peter H. (1966). "The Demand for International Travel by the United States and Canada."
International Economics Review.
Kwack, Sung Y. (1971). "Effects of Income and Prices on Travel Spending Abroad, 1960 III-
1967 IV." International Economics Review.
Orrenius, Pia M. (2003). "U.S. Immigration and Economic Growth: Putting Policy on Hold,"
Southwest Economy, Federal Reserve Bank of Dallas, November/December.
Passel, Jeffrey S. (2005). “Unauthorized Migrants: Numbers and Characteristics”, Pew Hispanic
Center White Paper, June 14.
Rhomberg, R.R. and L. Boissonneault (1964). "Effects of income and price changes on the U.S.
balance of payments," IMF Staff Papers.
Rose, Thom J. (2004). "U.S. visas: Applications down, prices up." United Press International.
April 15.
Siskin, Alison (2004). “Visa Waiver Program.” CRS Report for Congress, December.
Vilasuso, Jon and Fredric C. Menz. (1998). "Domestic Price, (Expected) Foreign Price, and
Travel Spending by Canadians in the United States." The Canadian Journal of
Economics.
31
Table 1: Participation in the Visa Waiver Program
Month Year Month Year
(1) United Kingdom July 1988
(2) Japan December 1988
(3) France July 1989
(4) Switzerland July 1989
(5) Germany July 1989
(6) Sweden July 1989
(7) Italy July 1989
(8) Netherlands July 1989
(9) Andorra October 1991
(10) Austria October 1991
(11) Belgium October 1991
(12) Denmark October 1991
(13) Finland October 1991
(14) Iceland October 1991
(15) Lichtenstein October 1991
(16) Luxembourg October 1991
(17) Monaco October 1991
(18) New Zealand October 1991
(19) Norway October 1991
(20) San Marino October 1991
(21) Spain October 1991
(22) Brunei July 1993
(23) Ireland April 1995
(24) Argentina July 1996 February 2002
(25) Australia July 1996
(26) Slovenia September 1997
(27) Portugal August 1999
(28) Singapore August 1999
(29) Uruguay August 1999 April 2003
Country
Date of Inclusion Date of Removal
32
Table 2: Summary Statistics of Non-immigrant Entries in the Sample
Fiscal Year 2000 Fiscal Year 2003 Decline
Aggregate Data 204 34,113,528 28,214,826 -17%
Full Sample 65 28,953,399 24,158,712 -17%
V
isa-Waiver Eligible and Canada 22 18,896,232 14,805,762 -22%
(1) Australia 593,246 552,916 -7%
(2) Austria 213,384 131,340 -38%
(3) Belgium 258,904 179,559 -31%
(4) Denmark 178,349 163,408 -8%
(5) Finland 118,369 86,742 -27%
(6) France 1,329,169 1,040,949 -22%
(7) Germany 2,146,442 1,444,665 -33%
(8) Iceland 31,803 25,229 -21%
(9) Ireland 405,583 372,137 -8%
(10) Italy 810,613 641,216 -21%
(11) Japan 5,259,703 3,593,469 -32%
(12) Luxembourg 12,913 7,640 -41%
(13) Netherlands 684,041 546,191 -20%
(14) New Zealand 200,147 204,219 2%
(15) Norway 166,300 142,935 -14%
(16) Portugal 114,701 84,436 -26%
(17) Singapore 119,632 81,919 -32%
(18) Slovenia 18,035 11,732 -35%
(19) Spain 466,168 430,070 -8%
(20) Sweden 377,000 257,899 -32%
(21) Switzerland 390,237 257,898 -34%
(22) United Kingdom 5,001,493 4,549,193 -9%
Non Visa-Waiver Eligible 40 5,043,926 4,506,576 -11%
(1) Belize 30859 30,859 26,587 -14%
(2) Botswana 2392 2,392 1,945 -19%
(3) Chile 211738 211,738 140,553 -34%
(4) Colombia 478142 478,142 389,768 -18%
(5) Costa Rica 173112 173,112 149,998 -13%
(6) Croatia 24925 24,925 19,960 -20%
(7) Cyprus 13345 13,345 9,554 -28%
(8) Czech Republic 52910 52,910 44,478 -16%
(9) Ecuador 138661 138,661 163,531 18%
(10) Estonia 8856 8,856 8,758 -1%
(11) Greece 79359 79,359 60,083 -24%
(12) Georgia 4869 4,869 5,290 9%
(13) Guatemala 183162 183,162 189,989 4%
(14) Hong Kong 129401 129,401 75,780 -41%
(15) Hungary 68696 68,969 45,020 -35%
(16) India 560110 560,110 559,805 0%
(17) Indonesia 97249 97,247 65,071 -33%
(18) Iran 28425 28,425 10,398 -63%
(19) Israel 357644 357,644 307,101 -14%
(20) Jamaica 277895 277,895 224,478 -19%
(21) Jordan 30631 30,631 21,214 -31%
(22) Korea 811951 811,951 845,272 4%
(23) Kyrgyzstan 2000 2,000 1,668 -17%
(24) Latvia 11733 11,733 10,494 -11%
(25) Lithuania 13747 13,747 15,811 15%
(26) Macau 1439 1,439 937 -35%
(27) Malaysia 95709 95,709 53,160 -44%
(28) Malta 9112 9,112 5,790 -36%
(29) Mauritius 3028 3,028 1,415 -53%
(4) Morocco 27590 27,590 18,021 -35%
(5) Peru 229307 229,307 224,542 -2%
(6) Philippines 281463 281,463 273,439 -3%
(7) Poland 147125 147,125 155,810 6%
(8) Romania 44162 44,162 49,722 13%
(9) Slovak Republic 20933 20,933 24,629 18%
(10) South Africa 132818 132,818 108,232 -19%
(11) Sri Lanka 17656 17,656 13,846 -22%
(12) Thailand 97560 97,560 74,195 -24%
(13) Tunisia 12885 12,885 4,181 -68%
(14) Turkey 131056 131,056 106,051 -19%
Excluded 3 5,013,241 4,846,374 -3%
(1) Argentina 548,798 244,477 -55%
(2) Canada 260,086 243,755 -6%
(3) Mexico 4204357 4,204,357 4,358,142 4%
Number of Non-Immigrant Entries
Number of
Countries
33
34
(1) (2) (3) (4)
Post-9/11 * Visa -0.186 -0.113 -0.086 0.037
Waiver Program (0.091)** (0.055)** (0.084) (0.053)
VWP 0.097 0.497 -0.005 0.716
(0.077) (0.470) (0.053) (0.590)
Real GDP Growth
in Home Country
Real Exchange
Rate Growth
Separate Time Tren
Weighted
Fixed Effects
y
Excluded Countries a
a
Countries
Observations
* Significant at 10%
Log Entrants
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial
correlation.
0.675 0.951 0.977 0.911
(0.499) (0.290)*** (0.321)*** (0.226)***
-0.243 -0.206 -0.259 -0.156
(0.081)*** (0.086)** (0.049)*** (0.057)***
ds N N Y Y
Y N Y N
Time Time Time Time
Country Country Country Countr
Canada Canada Canada Canad
Mexico Mexico Mexico Mexico
Argentina Argentina Argentina Argentin
62 62 62 62
4604 4604 4604 4604
, ** Significant at 5%, *** Significant at 1%
Table 3: Baseline Specification Regression Results
Table 4: Results Conditional on Domestic Demand
(1) (2) (3) (4) (5) (6) (7) (8)
Post-9/11 * Visa -0.172 -0.091 -0.078 0.044 -0.187 -0.102 -0.077 0.035
Waiver Program (0.087)* (0.056) (0.081) (0.057) (0.084)** (0.060)* (0.071) (0.064)
Real Growth in Home 0.498 0.137 0.682 -0.090 0.538 0.310 0.769 0.136
Travel Expenditures (0.412) (0.200) (0.335)** (0.222) (0.406) (0.179)* (0.325)** (0.205)
VWP 0.097 0.536 -0.025 0.780 0.082 0.510 -0.021 0.776
(0.082) (0.498) (0.042) (0.634) (0.083) (0.501) (0.041) (0.640)
Real GDP Growth 0.351 1.046 0.557 1.044
in Home Country (0.408) (0.296)*** (0.193)*** (0.241)***
Real Exchange -0.253 -0.169 -0.279 -0.145 -0.264 -0.228 -0.302 -0.208
Rate Growth (0.092)*** (0.053)*** (0.054)*** (0.036)*** (0.080)*** (0.048)*** (0.055)*** (0.041)***
Separate Time Trends N N Y Y N N Y Y
Weighted YNYNYNYN
Fixed Effects TimeTimeTimeTimeTimeTimeTimeTime
Country Country Country Country Country Country Country Country
Excluded Countries Canada Canada Canada Canada Canada Canada Canada Canada
Mexico Mexico Mexico Mexico Mexico Mexico Mexico Mexico
Argentina Argentina Argentina Argentina Argentina Argentina Argentina Argentina
Countries 5959595959595959
Observations 4223 4223 4223 4223 4223 4223 4223 4223
* Significant at 10%, ** Significant at 5%, *** Significant at 1%
Log Entrants
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial correlation.
35
36
VWP Non-VWP Log Entrants
(1) (2) (3) (4)
Portugal Czech Republic Post-9/11 * Visa -0.160 -0.017 -0.179 -0.038
Slovenia Greece Waiver Program (0.081)* (0.117) (0.089)* (0.183)
Spain Mauritius
New Zealand Korea VWP 0.162 0.978 0.138 1.412
Italy Malta (0.072)** (0.878) (0.099) (0.939)
Cyprus
Israel Real GDP Growth 2.753 3.271 2.725 2.566
in Home Country (1.054)** (1.826) (0.901)** (1.502)
Real Exchange 0.084 0.188 0.064 0.061
Rate Growth (0.201) (0.253) (0.179) (0.265)
Separate Time Trends N N Y Y
Weighted Y N Y N
Fixed Effects Time Time Time Time
Country Country Country Country
Countries 12 12 12 12
Observations 940 940 940 940
* Significant at 10%, ** Significant at 5%, *** Significant at 1%
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial correlation.
Table 5: Results Conditional on Country GDP
37
Coefficient Std. Error
Population 1995 5.61E-04 1.98E-04
Trade 1995 1.06E-08 1.45E-09
PPP-relative GDP 1995 5.27E-02 9.70E-04
Business share 2000 3.13E-01 2.81E-01
Ln_Distance -2.32E-01 3.71E-02
Propensity Score Probit
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial
correlation.
Panel B: Regression Results from Propensity-Score Restricted Dataset
Panel A: Propensity Score Determination (Probit)
VWP
Australia Botswana
Austria Chile Post-9/11 * Visa
Belgium Croatia Waiver Program
Denmark Cyprus
Finland Czech Republic VWP
France Estonia
Germany Georgia
Iceland Greece Real GDP Grow
Ireland Hong Kong in Home Country
Italy Israel
Netherlands Korea Real Exchange
New Zealand Kyrgzstan Rate Growth
Norway Latvia
Portugal Lithuania Separate Time T
Spain Macau
Sweden Malta Weighted
United Kingdom Mauritius
Slovak Republic Fixed Effects
Countries
Observations
* Significant at
Table 6: Regression Results Conditional on Propensity Score
Non-VWP Log Entrants
(1) (2) (3) (4)
-0.086 -0.024 -0.136 0.078
(0.117) (0.066) (0.098) (0.053)
0.067 0.058 -0.070 0.001
(0.076) (0.047) (0.051) (0.078)
th 1.783 0.955 1.598 0.852
(0.708)** (0.364)** (0.581) (0.341)**
-0.042 0.069 -0.130 0.103
(0.136) (0.219) (0.060) (0.198)
rends N N Y Y
Y N Y N
Time Time Time Time
Country Country Country Country
35353535
2736 2736 2736 2736
10%, ** Significant at 5%, *** Significant at 1%
Table 7: Regression Results Conditional on Business Travelers and View of U.S.
(1)(2)(3)(4)(5)(6)(7)(8)
Post-9/11 * Visa -0.118 -0.109 -0.036 0.045 -0.378 -0.262 -0.202 -0.031
Waiver Program (0.084) (0.054)** (0.066) (0.053) (0.098) (0.072)*** (0.108)* (0.070)
Post-9/11 * Share of 0.747 0.314 0.569 0.769
Business Travelers (0.518) (0.511) (0.255)** (0.354)**
Post-9/11 * View of U.S. -0.537 -0.205 -0.094 -0.219
(0.251)** (0.063)*** (0.141) (0.058)***
VWP 0.060 0.489 0.014 0.726
(0.076) (0.474) (0.058) (0.589)
Real GDP Growth 0.899 0.965 1.058 0.927 0.626 0.307 1.158 1.026
in Home Country (0.402)** (0.277) (0.327)*** (0.217)*** -0.685 (0.367) (0.413)** (0.321)***
Real Exchange -0.172 -0.196 -0.215 -0.147 -0.257 -0.147 -0.281 -0.150
Rate Growth (0.070)** (0.079) (0.037)*** (0.052)*** (0.077)*** (0.034)*** (0.071)*** (0.036)***
Separate Time Trends N N Y Y N N Y Y
Weighted YNYNYNYN
Fixed Effects Time Time Time Time Time Time Time Time
Country Country Country Country Country Country Country Country
Excluded Countries Canada Canada Canada Canada Canada Canada Canada Canada
Mexico Mexico Mexico Mexico Mexico Mexico Mexico Mexico
Argentina Argentina Argentina Argentina Argentina Argentina Argentina Argentina
Countries 62 59 59 59 17 17 17 17
Observations 4604 4223 4223 4223 1316 1316 1316 1316
* Significant at 10%, ** Significant at 5%, *** Significant at 1%
Log Entrants
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial correlation.
38
(1) (2) (3) (4) (5) (6) (7) (8)
Post-9/11 * Visa -0.211 -0.156 -0.119 -0.010 -0.214 -0.122 -0.125 -0.023
Waiver Program (0.092)** (0.058)*** (0.083) (0.053) (0.093)** (0.059)** (0.085) (0.066)
Post-9/11 * Share of -0.255 -0.255 -0.370 -0.268 -0.264 -0.233 -0.387 -0.276
Muslims in the Population (0.127)** (0.093)*** (0.103)*** (0.055)*** (0.121)** (0.092)** (0.106)*** (0.055)***
Post-9/11 * Visa Waiver Program 0.184 -1.338 0.322 0.448
* Share of Muslims in the Population (1.134) (0.744) (0.441) (0.653)
VWP 0.106 0.507 -0.008 0.713 0.100 0.552 -0.006 0.718
(0.075) (0.468) (0.049) (0.592) (0.083) (0.480) (0.053) (0.595)
Real GDP Growth 0.683 1.012 0.969 0.938 0.693 0.990 0.975 0.938
in Home Country (0.504) (0.286)*** (0.325)*** (0.212)*** (0.488) (0.279)*** (0.324)*** (0.211)***
Real Exchange -0.244 -0.162 -0.256 -0.125 -0.242 -0.159 -0.254 -0.126
Rate Growth (0.081)*** (0.063)** (0.052)*** (0.049)** (0.080)*** (0.067)** (0.049)*** (0.048)
Separate Time Trends NNYYNNYY
Weighted YNYNYNYN
Fixed Effects Time Time Time Time Time Time Time Time
Country Country Country Country Country Country Country Country
Excluded Countries Canada Canada Canada Canada Canada Canada Canada Canada
Mexico Mexico Mexico Mexico Mexico Mexico Mexico Mexico
Argentina Argentina Argentina Argentina Argentina Argentina Argentina Argentina
Countries 62 62 62 62 62 62 62 62
Observations 4606 4606 4606 4606 4606 4606 4606 4606
* Significant at 10%, ** Significant at 5%, *** Significant at 1%
Log Entrants
39
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial correlation.
Table 8: Regression Results Conditional on Muslim Population
Figure 1: Non-Immigrant Entries
Panel A: Annual Non-Immigrant Entries (Published)
20
22
24
26
28
30
32
34
36
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Millions
Fiscal Years Ending September 30
Panel B: Monthly Non-Immigrant Entries (Our Dataset)
1.5
1.8
2.0
2.3
2.5
2.8
3.0
3.3
3.5
Millions (Seasonally Adjusted Monthly Data)
1996 1997 1998 1999 2000 2001 2002 2003
40
Figure 2: Passengers on U.S. Domestic Non-Stop Segments
30
35
40
45
50
55
60
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
U.S. Domestic Non-Stop Passenger Emplanements, Millions
Source: Bureau of Transportation Statistics, U.S. Department of Transportation
41
Figure 3: Monthly Non-Immigrant Entries, by Participation in the Visa Waiver Program
Panel A: Levels
0.6
0.8
1.0
1.2
1.4
1.6
1.8
1996 1997 1998 1999 2000 2001 2002 2003
Millions (Seasonally Adjusted Monthly Data)
VWP Countries
(Excluding Canada)
Non-VWP Countries
(excluding Mexico)
Note: Constant panel from our dataset, excludes Argentina, Australia, Portugal, Singapore, Slovenia, and Uruguay
Panel B: Natural Log Levels
13.4
13.6
13.8
14.0
14.2
14.4
1996 1997 1998 1999 2000 2001 2002 2003
Natural Log Levels (Seasonally Adjusted Monthly Data)
VWP Countries
(Excluding Canada)
Non-VWP Countries
(excluding Mexico)
Note: Constant panel from our dataset, excludes Argentina, Australia, Portugal, Singapore, Slovenia, and Uruguay
42
Panel B: Separate Time Fixed Effects (Unweighted)
Panel A: Separate Time Fixed Effects, Baseline Case (Weighted)
43
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
1996 1997 1998 1999 2000 2001 2002 2003
9/11 Attacks
Major Combat
Operations in Iraq
Time Fixed Effects (Seasonally Adjusted and Demeaned)
VWP Countries
(Excluding Canada)
Non-VWP Countries
(excluding Mexico)
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
1996 1997 1998 1999 2000 2001 2002 2003
9/11 Attacks
Time Fixed Effects (Seasonally Adjusted and Demeaned)
VWP Countries
(Excluding Canada)
Non-VWP Countries
(excluding Mexico)
Figure 4: Separate Time Fixed Effects
Major Combat
Operations in Iraq
Table A1: Regression Results For Various Data Samples
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Post-9/11 * Visa -0.247 -0.100 0.013 0.018 -0.275 -0.105 -0.009 0.039 -0.155 -0.109 -0.061 0.015
Waiver Program (0.111)** (0.058)* (0.097) (0.072) (0.105)** (0.058)* (0.095) (0.052) (0.098) (0.054)** (0.086) (0.075)
VWP 0.277 0.447 0.185 0.508 0.054 0.472 -0.020 0.712 0.273 0.466 0.222 0.514
(0.128)** (0.350) (0.097) (0.347) (0.084) (0.468) (0.066) (0.586) (0.105)** (0.352) (0.100)** (0.348)
Real GDP Growth 0.634 0.920 0.989 -0.128 0.762 0.965 1.312 0.953 0.560 0.899 0.650 0.842
in Home Country (0.355)* (0.271)*** (0.435)** (0.230)*** (0.414)* (0.285)*** (0.365)*** (0.224)*** (0.417) (0.274)*** (0.372)* (0.232)***
Real Exchange -0.278 -0.195 -0.235 -0.128 -0.219 -0.197 -0.241 -0.157 -0.289 -0.202 -0.258 -0.127
Rate Growth (0.058)*** (0.087)** (0.050)*** (0.067)* (0.059)*** (0.089)** (0.048)*** (0.054)*** (0.069)*** (0.085)** (0.049)*** (0.069)*
Separate Time Trends N N Y Y N N Y Y N N Y Y
Weighted YNYNYNYNYNYN
Fixed Effects TimeTimeTimeTimeTimeTimeTimeTimeTimeTimeTimeTime
Country Country Country Country Country Country Country Country Country Country Country Country
Excluded Countries None None None None Argentina Argentina Argentina Argentina Canada Canada Canada Canada
Mexico Mexico Mexico Mexico
Countries 65 65 65 65 64 64 64 64 63 63 63 63
Observations 4856 4856 4856 4856 4772 4772 4772 4772 4688 4688 4688 4688
* Significant at 10%, ** Significant at 5%, *** Significant at 1%
Log Entrants
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial correlation.
44
Table A2: Regression Results For Various Weighting Schemes
(1) (2) (3) (4) (5) (6)
Post-9/11 * Visa -0.060 0.066 -0.247 0.080 -0.134 -0.079
Waiver Program (0.065) (0.050) (0.080)*** (0.050)* (0.106) (0.091)
VWP 0.419 0.561 0.165 0.169 0.081 0.087
(0.383) (0.489) (0.153) (0.203) (0.058) (0.045)*
Real GDP Growth 1.022 1.040 0.069 0.958 0.995 1.282
in Home Country (0.337)*** (0.242)*** (0.249) (0.201)*** (0.290)*** (0.212)***
Real Exchange -0.246 -0.161 -0.246 -0.191 -0.303 -0.292
Rate Growth (0.068)*** (0.056)*** (0.090)*** (0.028)*** (0.081)*** (0.068)***
Separate Time Trends N Y N Y N Y
Weighting Variable (All from 1995) Per-Capita GDP Per-Capita GDP Population Population Bilateral Trade (USD) Bilateral Trade (USD)
Fixed Effects Time Time Time Time Time Time
Country Country Country Country Country Country
Excluded Countries Canada Canada Canada Canada Canada Canada
Mexico Mexico Mexico Mexico Mexico Mexico
Argentina Argentina Argentina Argentina Argentina Argentina
Luxembourg
#
Luxembourg
#
Botswana
#
Botswana
#
Countries 62 62 62 62 60 60
Observations 4604 4604 4604 4604 4452 4452
* Significant at 10%, ** Significant at 5%, *** Significant at 1%
#
Missing 1995 Trade Data
Log Entrants
Notes: Standard errors are robust to general heteroskedasticity and are clustered by country to account for serial correlation.
45
