Journal of Urban Economics 61 (2007) 129–151

www.elsevier.com/locate/jue

Out of control: What can we learn from the end

of Massachusetts rent control?
David P. Sims ∗
Economics Department, Brigham Young University, 130 FOB, Provo, UT 84602, USA
Received 13 February 2006; revised 16 June 2006
Available online 22 August 2006

Abstract
This paper uses the sudden end of rent control in Massachusetts in 1995 to estimate the effects of rent
control. I examine Boston MSA data from the American Housing Survey years 1985–1998 to determine
how rent control affected the quantity, price and quality of rental housing. My results suggest rent control
had little effect on the construction of new housing but did encourage owners to shift units away from rental
status and reduced rents substantially. Rent control also led to deterioration in the quality of rental units, but
these effects appear to have been concentrated in smaller items of physical damage. I also examine spec-
ifications that allow rent control to affect rent levels both directly through controlled status and indirectly
through spillover effects from nearby rent controlled units. These estimates imply that rent control may
have small effects on the price of the non-controlled rental housing stock.
© 2006 Elsevier Inc. All rights reserved.

Keywords: Rent control; Maintenance; Housing supply; Spillover effects

1. Introduction

In a 1995 article Richard Arnott [3] called for “revisionism on rent control,” contending that
most economists hold traditional views about the effects of rent control that are founded upon
unrealistic models and scant empirical evidence. Indeed, few government policies have united
economists in opposition as effectively as rent control. A well known 1990 survey of econo-
mists [2] found that over ninety percent believed that rent control decreased the quantity and
quality of rental housing in an area. In contrast to this virtual unanimity of opinion, the work of

* Fax: +1 (801) 422 0194.

E-mail address: [email protected].

0094-1190/$ – see front matter © 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.jue.2006.06.004
130 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

Olsen [19] and Kutty [15] predicts that the effect of rent control on housing supply and rental
quality is theoretically ambiguous and must be empirically determined.1
Most recent empirical evidence on the effects of rent control in the US comes from stud-
ies of New York City.2 While New York City has had the most domestic experience with rent
control, measured in volume of rent controlled apartments, its rent controls consist of complex
overlapping sets of regulation regimes enacted at different points in time. This makes it atypical
of American cities that have experienced rent control and makes measurement of the effects of
any one set of rent controls difficult.
However, rent control policies have also played an important role in hundreds of American
cities from San Francisco to Boston. Many of these policies rose out of social unrest and tenant
activism in the late 1960s and early 1970s and remained the focus of animosity between activists
and landlords. In 1990, a group of homeless advocates led the seizure of an apartment building
in Cambridge that had been left vacant in violation of the city control ordinance. Local landlords
complained to the city council that with rents set below 150 dollars, the units were not worth
the trouble of renting and threatened to raise a private army to prevent further seizures by force.
Angered by the loss of property rights, these owners sought a political climate in which they
could end rent control.
The opportunity came in November 1994, when landlords succeeded in placing an initiative
on the Massachusetts ballot to ban rent control statewide. Though this initiative, known as Ques-
tion 9, passed statewide, it was overwhelmingly defeated in the three Massachusetts cities with
rent control, Boston, Brookline, and Cambridge. This externally imposed end to rent control
provides an opportunity to study the effects of rent control on housing units in Eastern Massa-
chusetts.
This paper examines the effects of rent control in Massachusetts on the willingness of owners
to rent housing units, on the rent and cost levels of renter occupied apartments, the maintenance
of those apartments, and length of tenancy. The results suggest that eliminating rent control has
little effect on increasing construction in the years immediately following the law change, but
that rent control induces owners to remove their units from the rental market. Additionally, rent
control leads to large rent decreases and small but significant decreases in the maintenance of
rental units. I also find evidence that rent control may lower the rent of non-controlled units in
controlled areas, possibly through spillover effects of decreased unit quality. Using my estimates
I calculate a rough measure of Massachusetts’ approximate welfare losses in due to rent control.
These results confirm the simple intuition that economists derive from very basic microeco-
nomic models; rent control artificially lowers price, decreases the quantity of rental housing
supplied and decreases unit quality. Although inefficiencies are inherent in any price control,
rent control is an opportunity to study a price control that is large in magnitude and has impor-
tant effects on the behavior of many tenants and landlords.
These findings also provide important evidence to policy makers. A decade after the abolition
of Massachusetts rent control the debate over its legacy continues. The Mayor of Boston, Thomas
Mennino, often lobbies for a return to rent control and blames the repeal of rent control laws for
the skyrocketing Boston rental rates of the late 1990s. Opponents argue that this price increase
was due to an economic boom and that renewed rent control will exacerbate the fundamental

1 Olsen [19] shows that modeling tenant maintenance makes the effect of rent control on housing quality ambiguous.
This also implies that the tenants’ expense of acquiring housing services may not be fully captured by the rent they pay.
2 Important studies of New York include Early [8], Glaeser and Luttmer [11], two studies by Gyourko and Linneman
[12,13], Moon and Stotsky [16], and Olsen [18].
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 131

problem of unit undersupply. This study provides numerical estimates to inform both sides of the
debate.
This study of rent control is also relevant to the policy process of other cities. From San
Francisco to Washington DC there is a need to assess the potential impact of further weakening
or completely repealing existing controls. Given the Supreme Court decision in Chevron v. Lingle
that reaffirmed earlier court decisions that rent control is not an unconstitutional taking of private
property, it seems likely that rent control will continue to appear on the policy agenda of large
cities in the future. This study provides a clearer empirical understanding of how rent control
impacts cities.
The remainder of the paper is organized as follows: Section 2 provides additional institutional
detail about rent control in Massachusetts and its repeal, Section 3 discusses the data, Section 4
presents the results and Section 5 concludes.

2. Rent control in Massachusetts

2.1. The laws

Of the Massachusetts cities that adopted rent control in the late 1960s and early 1970s, only
Boston (1969), Cambridge (1971), and Brookline (1970) maintained rent control ordinances past
the early 1980s into the period covered by this study (1985–1998). Though the laws in each city
varied in some details, all rent control policies shared four common elements.
The first was a centralized board or commission which was empowered to set maximum
rents. There were few upward adjustments made to rents in the 1970s and early 1980s despite
high inflation. However, in the late 1980s rent control boards would often approve general rent
increases for all controlled apartments in the city. Boston made several general rent adjustments
of about 5%, Brookline gave a usual annual increase of 4%, and Cambridge indexed increases
to a multiple (0.85) of CPI growth. Other increases had to be approved on an individual basis,
requiring the landlord to provide proof of an operating cost increase such as a rise in property
taxes or utility rates.3
The second common element was a set of laws to reduce condominium conversions that might
take apartments out of the rental stock. In Cambridge, for example, condominium conversions
required the express approval of the rent control board. In Boston, the Bar Association [5] noted
that owners of buildings destined for condominium conversion were also required to give tenants
up to three years advance notice before conversion, help the tenants find new housing, and pay a
severance fee. These regulations made it very difficult to remove a controlled unit by converting
it into a condominium.
A third set of regulations concerned tenant protection. These forbade the eviction of tenants
without approval of the board. A short list of grounds for eviction was codified in the laws, and
landlords faced the burden of proving violations. Another regulation set monetary punishments
for landlords who failed to maintain the provision of essential services such as heat and running
water to their units. Most controversially, Cambridge forbade landlords from leaving controlled
units vacant for more than three months.4

3 Even then the process could be excruciating. In 1988–1989 the Cambridge board [7] reported the median time for
the adjudicating a rent increase for fair operating costs was 171 days. Even with general rate increases almost 10% of
controlled units in Cambridge had a set rent of 200 dollars or less in 1989.
4 There is little evidence that the Cambridge ordinance banning vacancies was ever actively enforced.
132 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

The final common element was a system for removing units from further control. Each city
exempted newly constructed housing and housing units that were completely remodeled, pro-
vided a certain amount of money was spent updating the units.5 Boston and Cambridge also
exempted owner-occupied 2 and 3 family homes from rent control.
Boston and Brookline had forms of vacancy decontrol during this period. Boston’s decontrols
were adopted in 1984, just before the period covered by this study. These regulations allowed
a rental unit to leave active control once it was completely vacated by its present tenants. It
then passed into a state of passive control, still subject to the removal and eviction protections.
Landlords were free to raise rent on a yearly basis, but tenants in such apartments could ap-
peal to the rent control board to override unfair rent increases. In Brookline [6], a simpler
rent decontrol system, adopted in 1991, exempted most vacated apartments from all rent con-
trols.
As a result, a relatively small percentage of rental housing in the three communities was
covered by active rent control. A 1991 HUD report [23] estimated that 38% of the rental units
in Cambridge and 49% of the rental units in Brookline (about 19,000 units between the two
cities) were rent controlled while 50% of the units in Boston fell under either active rent control
or vacancy decontrol. William Galvin [10] noted that at the time of repeal only about 22,000
Boston housing units (about 14% of total rental housing) were actually under active rent control
while approximately 60,000 others qualified for vacancy decontrol. Thus only about 20% of the
rental housing stock (about 41,000 units) of the cities with rent control was actively controlled at
this time.

2.2. The repeal

Throughout the early 1990s the fight to end rent control was led by the Small Property Owners
Association (SPOA). Massachusetts cities with rent control had a majority of tenant residents as
well as government officials with a favorable attitude toward rent control. Thus SPOA had little
success in repealing rent control laws through local action. Finally, in 1994, they changed tactics
and proposed a statewide ban of rent control in the form of a ballot initiative, Question 9. The
ensuing political campaign was bitterly contested on both sides, with a number of tenant activist
groups organizing opposition to the proposed law.
In the November election the voters approved the Question by a narrow margin (51%
favored it). After failing to obtain judicial intervention, tenant advocates sought help from
the state legislature. They argued the end of rent control on January 1, 1995 would be
marked by wholesale evictions of the poor and elderly, and a massive rise in homeless-
ness. Finally, the governor brokered a compromise whereby tenants that met certain age or
poverty guidelines could retain their controlled unit status for a 1–2 year transition period.
Thus, most units were immediately decontrolled, but a few remained in control until January
1997.6

5 In Boston the required investment ranged from $15,000 to over $35,000 per unit depending on the age of the unit,
and the year of renovation.
6 About 3000 people were granted these extension waivers, supporting the case of rent control opponents that most
rent controlled tenants were not poor or elderly.
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 133

3. Data & estimation

3.1. Data description

The primary data for this study come from the American Housing Survey–Metropolitan Sam-
ple (AHS-MS) for the Boston MSA for the years 1985, 1989, 1993, and 1998 [21]. The AHS-MS
interviews a sample of households drawn from census long forms in several cities across the
United States on a rotating basis. The unit of observation is the housing unit, not the inhabitant.
This survey is particularly useful because it asks residents a wide variety of questions about the
rent, maintenance, and physical characteristics of their units and provides a wealth of information
on the inhabitants. There are, however, disadvantages to using this data.
One primary limitation is the level of geographical identification available to the researcher.
Due to confidentiality concerns, the smallest identifiable geographical unit in the Public Use
files of the AHS-MS is the zone. Though vaguely defined as, “a roughly homogeneous region of
greater than 100,000 population” by the Census Bureau, in practice a zone in the Boston MSA
corresponds to a small group of towns or cities. An exception is the city of Boston where a zone
corresponds to a group of neighborhoods.
Because it is impossible to make more precise geographical distinctions than the zone, I treat
entire zones as controlled or non-controlled even though two zones are only partially subject to
rent control. These are zone 112 which encompasses Cambridge and Somerville and zone 110
which includes Brookline and Newton. In both cases the geographical proximity and similarity
of the cities involved supports the idea that they comprise a rental market. Nevertheless, the effect
of being in a rent controlled zone, explored by this paper, is not precisely the same as the effect
of being in a rent controlled city.
Furthermore some of the Boston MSA zones in the survey changed geographical boundaries
as the composition and extent of the MSA changed over time. A list of the 1990 census tracts
within each zone allows me to construct zone boundaries for constant geography zones in the
1985–1998 time period. I limit my sample to these constant boundary zones which include almost
all of the interior suburbs surrounding Boston. In practice this excludes many outlying regions
of the Boston MSA from the sample. However, the excluded areas comprise the portion of the
MSA least comparable with the rent controlled areas. Figure 1 provides a map of the zones in
the Boston MSA used in this study as well as a list of the towns or cities that comprise each of
the 21 zones. Figure 2 shows the zone boundaries and lists the neighborhoods that make up the
zones in the city of Boston.
In addition to geographical limitations there are other potential difficulties with the AHS-MS.
For the 1998 survey a completely new sample for the Boston area was drawn from the 1990
census. This change prevents the observation of individual units as they pass out of rent control.
However, the new sample does include some information about each unit’s status during the
1990 census, particularly whether it was owned or rented in 1990, which allows me to examine
how the tenure status of particular units changed with the end of rent control. In 1998 the survey
questionnaire rephrased some questions from previous survey years and was administered for the
first time through computer assisted interviewing, which may have affected the comparability of
responses over time.7 Most notably, certain items about the external condition of the unit that

7 Because I compare zones that were decontrolled with zones that remained control free, survey changes will only bias
my results if they have a differential effect on these two types of zones.
134 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

Zones 1–5 Zone 110 Zone 115 Salem Westwood

Boston Brookline Nahant Beverly Norwood
Newton N. Reading Manchester Sharon
Zone 107 Lynn Rockport Canton
Bedford Zone 111 Lynnfield Gloucester
Concord Arlington Reading Zone 121
Lincoln Belmont Saugus Zone 118 Ashland
Sudbury Watertown Wakefield Milton Framingham
Wayland Wilmington Quincy Natick
Wellesley Zone 112 Sherborn
Weston Cambridge Zone 116 Zone 119
Somerville Swampscott Braintree Zone 122
Zone 108 Peabody Holbrook Blackstone
Burlington Zone 113 Danvers Randolph Holliston
Lexington Malden Middleton Weymouth Hopedale
Waltham Medford Topsfield Hopkinton
Hamilton Zone 120 Marlborough
Zone 109 Zone 114 Wenham Dedham Mendon
Melrose Chelsea Essex Needham Milford
Stoneham Everett Ipswich Dover Millville
Winchester Revere Rowley Medfield Southborough
Woburn Winthrop Millis Upton
Zone 117 Norfolk
Marblehead Walpole

Fig. 1. AHS-MS geography constant zones.

 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 135

Zone 1 Beacon Hill South Boston (part)

Allston Downtown Roxbury (part)
Brighton North End N. Dorchester Zone 5
Roslindale South End Harbor Islands Neponsett
Jamaica Plain Chinatown Mattapan
Zone 4 West Roxbury
Zone 2 Zone 3 Mission Hill Hyde Park
Back Bay Charlestown Roxbury (part)
Fenway East Boston Dorchester

Source: author’s calculations based on Census Bureau data.

Fig. 2. Boston city zones AHS-MS.

136 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

were previously answered by the surveyor were now asked to the tenant. I have excluded these
items from my analysis.
I examine two different samples of Boston AHS data. The first sample includes all housing
units in each of the four survey years that were properly interviewed, no matter the tenure of
their residents. The second is a sub-sample that includes only the units in which the resident is
a tenant. This renters sample also excludes public housing units and housing units where the
tenant pays a non-monetary rent. Table 1 provides descriptive statistics for both samples, as well
as a comparison of the characteristics of rental units in zones under rent control.8 The study also
uses data on building permits issued by the localities in Eastern Massachusetts provided by the
Department of Housing and Urban Development [22].
The first row of Table 1 highlights an interesting feature of the data; residents appear to un-
derreport rent control. Administrative sources indicate nearly 20% of rental housing in the rent
controlled cities was under active rent control while only 12.8% of survey respondents in the
corresponding zones reported living in a rent controlled unit. Accounting for the geography dif-
ference between cities and AHS zones and the different weighting of units in the samples explains
about one third of this difference. The remainder is likely due to some combination of sampling
variation and tenant misreporting. Underreporting of rent control has previously been mentioned
in connection with the New York City Housing and Vacancy Survey. In practice underreporting
would not confound my estimates of the effect of being in a rent controlled zone, but it does
affect their interpretation. The possible underreporting, however, is an important consideration
in trying to assess the specific impact of rent control on controlled units.

3.2. Empirical strategy

This study seeks to identify the effect of rent control on housing in controlled zones. Specif-
ically, I examine the effects of rent control on several housing unit characteristics including the
quantity of rental housing supplied, the level of rent and housing costs, unit quality, and length
of renter tenure.
To identify the effects of rent control, I compare the change in average outcomes of interest
for units in zones that experienced rent decontrol and those in zones that did not change status.
This basic difference-in-differences strategy can be implemented in a regression framework and
modified to include observation level controls and zone or year fixed effects. Consider an example
where unit rent (P ) is the outcome of interest. The estimating equation:
Pij t = δDj t + φj + γt + Xij t θ + εij t (1)
represents the rent of unit i in zone j at time t as a linear function of whether the unit receives the
treatment of rent decontrol Dj t , unit characteristics Xij t , zone fixed effects φj , year effects γt ,
and an error term εij t . The treatment indicator equals one for all units in 1998 in zones which
were previously subject to rent control. If the included controls are sufficient to account for
other characteristics that influence rent then Least Squares estimates of δ may be interpreted as

8 Eight percent of overall rental units were single family structures while only one percent of rental units were single
family structures under rent control. However, most one unit structures were never controlled. According to Vanden-
broucke [24], this difference is an artifact of a known issue in the 1998 metropolitan sample whereby too many units
are indicated as single family and too few as part of 2–4 family structures. My estimates of the effects of being in a rent
controlled zone are robust to both omitting number of units in structure as a control and combining single family units
with 2–4 family units in a single categorical control.
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 137

Table 1
Descriptive statistics
All units All renters Controlled area
Rent controlled 0.046 0.128
(0.084) (0.093)
Rent (1998 dollars) 486.559 398.181
(293.283) (239.274)
Monthly housing cost 755.705 617.99 547.534
(521.244) (318.159) (282.558)
Condominium 0.075 0.053 0.046
(0.263) (0.224) (0.209)
Length of stay 11.859 5.987 6.839
(13.304) (8.594) (9.711)
Number of bedrooms 2.46 1.829 1.822
(1.189) (0.991) (1.004)
Number of other rooms 3.078 2.424 2.394
(1.258) (0.862) (0.791)
Year built 1959 1938 1933
(21.6) (24.5) (23.5)
Single family 0.419 0.080 0.014
(0.493) (0.271) (0.119)
2–4 family 0.298 0.431 0.433
(0.458) (0.495) (0.496)
5–12 unit building 0.102 0.189 0.211
(0.303) (0.392) (0.408)
13–25 unit building 0.043 0.080 0.110
(0.203) (0.271) (0.313)
26–50 unit building 0.033 0.056 0.065
(0.179) (0.230) (0.246)
51+ unit building 0.067 0.121 0.131
(0.250) (0.326) (0.338)
Unit is gas heated 0.399 0.395 0.362
(0.490) (0.489) (0.481)
Unit is oil heated 0.470 0.413 0.457
(0.499) (0.492) (0.498)
Unit is electrically heated 0.121 0.181 0.159
(0.326) (0.385) (0.366)
Other source of heat 0.010 0.011 0.022
(0.101) (0.103) (0.148)
Central air in unit 0.125 0.120 0.070
(0.331) (0.300) (0.256)
Off street parking included 0.451 0.542 0.322
(0.498) (0.498) (0.467)
Owner/manager present 0.140 0.287 0.325
(0.347) (0.453) (0.468)
Gas included in rent 0.118 0.222 0.234
(0.323) (0.415) (0.424)
Electricity in rent 0.097 0.196 0.202
(0.297) (0.397) (0.402)
Oil/coal/kerosene in rent 0.143 0.270 0.301
(0.350) (0.444) (0.459)
Sanitation in rent 0.776 0.860 0.954
(0.417) (0.347) (0.209)
(continued on next page)
138 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

Table 1 (continued)
All units All renters Controlled area
Maintenance problem 0.354 0.350 0.360
(0.478) (0.477) (0.480)
Number of problems 0.478 0.526 0.574
(0.809) (0.934) (1.016)
Chronic physical damage 0.084 0.114 0.129
(0.278) (0.318) (0.335)
N= 10,512 3590 1290
Notes. The table presents the author’s calculations based on AHS-MS Boston 1985, 1989, 1993, 1998 surveys. Standard
errors are listed in parentheses below variable means. The dataset excludes units that were not properly interviewed and
those in zones geographically redefined between surveys. Columns (2)–(3) further exclude units where rent payment was
not in cash. Rent and housing cost are measured in constant 1998 dollars.

the causal effect of decontrolling a zone on the rent of units in the zone. The key identifying
assumption is that the assignment of rent decontrol, conditional on included control variables, is
uncorrelated with the error term.
In addition to estimating the effect of rent decontrol on housing characteristics, I seek to
measure how it affected the quantity of rental units supplied. There are two primary ways in
which rent controls could distort the quantity of rental housing supplied. Rent control might affect
the extensive quantity of housing units supplied, that is the number of housing units available in
total, or it might affect the intensive margin of rental housing units, the percentage of all housing
units that are rented.
A lack of explicit data about the number of housing units in each zone during survey years
makes direct estimation of the effect of rent decontrol on housing construction difficult. I adopt
two approaches to overcome this difficulty. First, I look at the number of units covered by con-
struction building permits in Eastern Massachusetts during this time period. Second, I use the
sampling weights in the AHS-MS dataset to obtain a rough approximation of the number of
units in each zone for a particular year. For each zone I construct a variable that represents the
fraction of Boston MSA units in that zone in a given year. I regress that measure on zone aver-
age housing characteristics and rent control status. In this way I hope to measure the growth of
decontrolled zones relative to the MSA as a whole.
To estimate movement on the intensive margin, I first ask how rent control changes the prob-
ability an existing housing unit is rented. The statistical model:
Rij t = δDj t + φj + γt + Xij t θ + εij t (2)
estimated as a linear probability model with Rij t a dummy variable equal to one if the unit is a
rental and zero otherwise posits that
Pr(Rij t = 1 | φ, γ , D, X) = δDj t + φj + γt + Xij t θ. (3)

Thus, the resulting δ̂ approximates the marginal effect of rent decontrol on the probability a
unit is rented.9

9 I use linear probability models throughout the paper when the dependent variable is dichotomous. The resulting
coefficient estimates though inefficient are easily interpretable as approximating marginal effects and do not require
distributional assumptions. The main results of this paper are robust to the use of probit models. I report heteroskedasticity
robust standard errors.
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 139

The inclusion of some 1990 census data for units in the 1998 survey provides another way to
approach the issue. I use the census variables to identify a sample of all units that were owner
occupied in 1990. I then mark which units changed to renter occupancy between 1990 and 1998.
Using this sample, I estimate a linear probability model where the dependent variable is set equal
to one if the unit switched tenure status and zero otherwise. The explanatory variables include
whether the unit was in an area that was decontrolled and other unit characteristics as measured
in 1990.
Looking only at zone level average effects of rent control may mask the channel through
which they operate. Rent control might affect rent levels in a zone by exclusively suppressing the
rent of controlled units. Alternatively, the presence of controlled units may influence the rent of
non-controlled rental housing. To test for the existence of such spillover effects I estimate:

Pij t = φj + γt + ηRCij t + λPCT j t + Xij t π + νij t (4)

where RCij t is a dummy variable equal to one if the unit is controlled and equal to zero otherwise
and PCT j t gives the percentage of units in zone j that are controlled at time t.
The primary obstacle to identification of the effect of rent control on controlled units is the
possibility of omitted variable bias. Because rent control status is not randomly assigned to hous-
ing units, it may be correlated with unobserved unit characteristics, leading to bias in Least
Squares estimates of η. Furthermore, Acemoglu and Angrist [1] show that when a regressor
such as PCT j t is an average of another regressor (RCij t ), Least Squares Estimation is likely to
indicate the presence of spillover effects even when none exist.
An instrumental variables strategy that treats both controlled status and the percentage of
controlled units in a zone as endogenous can in theory produce consistent estimates of individual
unit effects and spillover effects of rent control. If there are no spillover effects (λ = 0), Eq. (4)
contains only one endogenous right hand side variable and a single instrumental variable is suf-
ficient for identification. A suitable instrument for this problem needs to be correlated with the
rent controlled status of housing units while not affecting the outcome of interest except through
its effect on rent control. My primary instrument is the interaction of an indicator variable set
equal to one for all units in a zone that was ever under control during the study period and a time
dummy indicating an observation from 1998.
However, in the presence of spillover effects multiple instruments are necessary to iden-
tify both direct and spillover coefficients. I construct additional instruments using institutional
details about the legal requirements for rent controlled units. For example, because owner-
occupied 2 family houses are excluded from control, the interaction of a 2-family house indicator,
an owner-occupied premises indicator, a pre-treatment year indicator, and a controlled zones in-
dicator provides a variable that is correlated with rent controlled status. The relevant exclusion
restriction imposed by this instrument is that the main effect of a two family owner occupied
house on rent is identical in controlled and uncontrolled zones. In a similar manner, I construct
other instruments that take advantage of clauses in the rent control laws exempting new construc-
tion and single family homes, by interacting dummy variables for those characteristics with time
and controlled zone indicators.
140 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

4. Results

4.1. Quantity supplied

Figure 3 shows the number of multi-family housing units for which building permits were
granted in Boston, Brookline and Cambridge in each year from 1980–2000, and illustrates the
need to avoid deceptive endpoints in a time series. Opponents of rent control in Massachusetts
often cite the permit data from 1990–2000 as evidence that rent control led to an undersupply
of rental housing. Indeed, looking at only those years it appears that the number of permits
granted rose markedly after the end of rent control. However, once 1980–1990 are added, the
low permit years of the early 1990s appear to be the anomaly.10 Of course, the permit graphs are
not definitive statements about unit supply, since they do not count actual construction activity,
do not account for demolition and abandonment, and fail to consider a condominium market
which primarily builds units for owner occupancy.
Table 2 provides more formal estimates of the effect of rent decontrol on extensive quantity
supplied, as well as intensive quantity supplied and the number of condominium units. The first
two columns address the effect of rent decontrol on the total quantity of housing units supplied
in a zone. The unit of observation is the zone. The extensive quantity measure is the fraction of

The figure shows the number of units approved under building permits in Boston, Brookline and Cambridge from 1980–
2000. Source: Department of Housing and Urban Development building permits database.

Fig. 3. Building permits for controlled cities 1980–2000.

10 Simple Difference-in-Differences analyses indicates that the end of rent control did not increase building permit
activity in formerly controlled areas relative to the rest of the Boston MSA or the issue of multifamily permits relative to
single family permits within the formerly rent controlled cities.
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 141

the MSA’s housing units in that year in each zone as calculated from the sampling weights in
the dataset. A zone that grows at the same rate as the MSA will retain the same fraction over
time. The regression controls for zone average housing characteristics. Throughout the paper the
reported zone fixed effects are relative to the excluded zone 114, which includes a number of
areas just outside Boston.
Table 2 indicates that being in a decontrolled zone leads to an increase of about 0.2 percentage
points in the relative quantity of housing supplied in that zone. This small effect, however, is

Table 2
Effects of rent decontrol on the quantity of housing supplied
Dependent variable Extensive supply Intensive supply Condominiums
Fraction of units in Probability a given Probability Unit has
the MSA unit is rented condo status
(1) (2) (3) (4) (5) (6)
Decontrol Effects
Ever controlled zone ∗ 1998 0.0018 0.0608 0.0812
(0.0038) (0.0237) (0.0090)
Boston ∗ 1998 0.0080 0.0829 0.0867
(0.0045) (0.0299) (0.0105)
Other controlled zone ∗ 1998 −0.0005 0.0170 0.0700
(0.0035) (0.0312) (0.0084)
Zone Effects
Allston–Brighton–Roslindale −0.0117 −0.0134 0.0469 0.0418 0.0169 0.0156
Jamaica Plain (0.0075) (0.0066) (0.0243) (0.0264) (0.0085) (0.0082)
Back Bay–Fenway– −0.0220 −0.0287 0.0067 0.0014 0.0399 0.0385
North End–South End (0.0111) (0.0102) (0.0260) (0.0277) (0.0126) (0.0121)
Charlestown–E. Boston– −0.0058 −0.0092 0.0260 0.0198 −0.0246 −0.0262
S. Boston–N. Dorchester (0.0067) (0.0061) (0.0267) (0.0239) (0.0077) (0.0082)
Mission Hill–Roxbury (part) −0.0333 −0.0372 0.1219 0.1169 −0.0446 −0.0458
Dorchester (0.0075) (0.0069) (0.0420) (0.0388) (0.0114) (0.0119)
Mattapan–W. Roxbury −0.0219 −0.0208 0.0174 0.0119 −0.0421 −0.0435
Hyde Park (0.0106) (0.0093) (0.0264) (0.0258) (0.0144) (0.0144)
Brookline–Newton −0.0088 −0.0077 0.0109 0.0225 0.0519 0.0548
(0.0074) (0.0065) (0.0291) (0.0236) (0.0097) (0.0099)
Cambridge–Somerville 0.0051 0.0036 0.0234 0.0337 −0.0037 −0.0011
(0.0046) (0.0041) (0.0180) (0.0189) (0.0082) (0.0081)
Year Effects
1989 0.0437 0.0428 0.0212 0.0212
(0.1395) (0.1397) (0.0160) (0.0162)
1993 0.1500 0.1494 0.2729 0.2728
(0.1223) (0.1225) (0.0938) (0.0938)
1998 0.0489 0.0485 0.1781 0.1776
(0.0828) (0.0831) (0.1024) (0.1024)
N= 84 84 10,480 10,480 10,512 10,512
Unit of observation Zone Zone Unit Unit Unit Unit
Notes. Columns (1)–(2) are zone level regressions with the fraction of the units in the MSA in each particular zone,
calculated from sampling weights, as the dependent variable. Columns (3)–(6) report estimates of linear probability
models with a dummy variable for rental tenure (3)–(4) or condo status (5)–(6) as the dependent variable. Regressions in
columns (3)–(6) include quadratics and cubics for number of bedrooms and other rooms, categorical age and number of
building units, age of building interactions with year, and controls for type of heat and presence of central air as well as
zone and year fixed effects. Columns (1)–(2) include zone level averages of the controls and include zone fixed effects.
Standard errors are corrected for clustering on the zone year level.
142 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

not statistically different from zero. When Boston and the other decontrolled zones are treated
separately Boston has a positive growth in quantity supplied due to rent decontrol that is not
shared by the zones of Cambridge and Brookline. However, the standard error of this estimate
remains large. These results suggest that in the short term the end of rent control had little effect
on the construction of new housing.
In contrast, columns (3)–(6) provide evidence that rent control decreases the available number
of rental units and the number of condominium units in a zone. Columns (3)–(4) report the
results of estimating Eq. (2) as a linear probability model. The unit of observation is the housing
unit, with the dependant variable set equal to one if the unit is a rental and zero otherwise.
Because a unit’s characteristics may affect its suitability for rental, all regressions flexibly control
for the number of bedrooms, number of rooms, age, type of building, and physical equipment.
Reported standard errors are corrected for clustering at the zone year level.11 The coefficients in
column (3) indicate that the end of rent control is associated with a 6 percentage point increase
in the probability of a unit being a rental. In regressions that consider condominium status in a
similar fashion, rent decontrol is associated with an 8 percentage point increase in the probability
of a unit being a condominium.
Though condominium units are often thought of as exclusively owner occupied, many con-
dominiums in the Boston area were rented during this time period. Thus, a straightforward
interpretation of the results is that rent control reduced condominium conversions in an attempt
to stop units from moving to owner occupancy. At the same time, units that already had con-
dominium status were less likely to be rented out as that would subject them to rent control.
Additionally, because the ubiquitous 2–3 family houses of New England were exempt from rent
control if one unit was owner occupied, the rent control laws provided an incentive for fewer of
those units to remain rentals.
Table 3 provides additional evidence that rent control caused units to be removed from rental
markets. It explores whether units in zones that experienced rent decontrol were more likely

Table 3
Transition of unit tenure 1990–1998
Outcome Transition owner–renter Transition renter–owner
(1) (2) (3) (4) (5) (6)
Zone decontrolled in 1995 0.119 0.082 0.071 −0.036 −0.024 −0.024
(0.017) (0.018) (0.018) (0.018) (0.018) (0.018)
Control for number of rooms no yes yes no yes yes
Control for categorical 1990 rent no no no no yes yes
Control for categorical 1990 value no yes yes no no no
Interaction of rooms/rent or value no no yes no no yes
N= 1875 1875 1875 1648 1648 1648
Notes. The reported coefficients are from linear probability models with outcome dummies = 1 if the unit changed tenure
status from 1990 to 1998. The sample in columns (1)–(3) is all units from the 1998 survey that were owner occupied
in 1990. The sample in (4)–(6) is all 1998 units that were renter occupied in 1990. Categorical controls reflect value
intervals provided in the data.

11 This provides conservative standard errors when using individual unit data with an aggregated regressor of interest,
such as my interaction of controlled zone and the year 1998. The downward bias of standard errors in this circumstance
was noted by Moulton [17]. Wooldridge [25] provides a more detailed discussion. Cluster corrected standard errors are
also used in Tables 4–6 and 8.
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 143

to switch from owner to tenant occupancy after rent control ended than units in other zones.
The sample used in the first three columns of the table includes all units from the 1998 Boston
AHS-MS that were listed as owner occupied in the 1990 census. The dependent variable is a
dummy equal to one if the unit switched to renter occupancy by 1998. The regressor of inter-
est is a dummy variable indicating a unit’s zone was decontrolled by the 1994 ballot initiative.
Columns (2) and (3) control for rent, value and unit size as indicated in the table.
The significant, positive coefficient estimates suggest that units in decontrolled zones are
about 7 percentage points more likely to become rental units after the end of rent control than
units in zones that were never controlled. Columns (4)–(6) provide a specification check by
demonstrating that rent decontrol has no significant effect on the transition of units in the oppo-
site direction, from renter to owner occupancy.
In summary, there is weak evidence that rent control affected the extensive quantity of housing
units supplied in Boston, but much stronger evidence that rent control led owners to shift units
away from renting. The 6–7 percentage point change in rental probability between controlled
and uncontrolled zones may seem small, but when applied to all three cities it implies that rent
control kept thousands of units off the market.

4.2. Rent, maintenance, and tenure

If rent control operates by capping rents below market rates, it seems sensible that controls
would have negative rent effects. This section measures the magnitude of these effects. Addition-
ally it presents findings about the effects of rent control on unit maintenance and the length of
renter tenure. Table 4 presents results from the estimation of Eq. (1) using a variety of outcome
variables.12 Column (1) measures the effect of being in a decontrolled zone on rent levels. It indi-
cates that the end of rent control is associated with an 84 dollar jump in rent.13 Additionally, the
year effect coefficients show that rents increased at an increasing rate throughout the study pe-
riod. Column (2) uses a more comprehensive measure of housing costs as the dependent variable.
It adds is the cost of utilities to the rent. The results imply that ending rent control causes housing
costs to jump by 64 dollars. The smaller coefficient on cost makes sense as rent control provides
strong incentives to landlords not to include any utilities in the rent, a practice they might forgo
when controls are lifted. Both the cost and rent coefficients are significantly different from zero.
Ideally, to estimate the effect of rent control on unit maintenance, I would use a full history
of the maintenance expenditures by the landlord over the life of the building. Unfortunately, this
data is not available. The rent control literature traditionally addresses this problem by looking
at the physical condition of rental units rather than maintenance spending.
I adopt this practice in columns (3)–(5), which demonstrate how the end of rent control af-
fected various measures of unit condition. Column (3) is a linear probability model where the
dependent variable takes on a value of 1 if the unit experienced a major maintenance problem
in the last season. These problems include plumbing and heating failures as well as pipe leaks,
wiring shorts or other electrical problems. The results indicate that decontrol reduces the proba-
bility of a unit experiencing such problems, but the coefficient is not significantly different from
zero. The dependent variable in column (4) is the number of these major problems experienced
by units. Again, the estimated effect is negative but insignificant.

12 Tables 4–6, and Table 8 report weighted least squares regressions that account for sampling procedures in the survey
design. The weight for a unit equals the inverse probability that it was sampled.
13 All rent and cost data are in 1998 dollars.
144 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

Table 4
Effects of rent decontrol on zone housing characteristics
Dependent variable rent cost severe # probs chronic length
(1) (2) (3) (4) (5) (6)
Decontrol Effects
Ever controlled zone ∗ 1998 83.78 64.25 −0.02 −0.11 −0.05 −1.84
(20.63) (24.73) (0.05) (0.08) (0.02) (0.59)
Zone Effects—Controlled Zones
Allston–Brighton–Roslindale 100.50 109.27 0.08 0.07 0.02 0.61
Jamaica Plain (26.86) (32.95) (0.04) (0.12) (0.03) (0.78)
Back Bay–Fenway–Downtown 200.14 199.51 0.04 0.02 0.00 1.11
North End–South End (25.37) (31.36) (0.05) (0.11) (0.03) (0.87)
Charlestown–E. Boston– −75.25 −77.30 −0.04 −0.19 −0.06 3.23
S. Boston–N. Dorchester (32.35) (33.94) (0.05) (0.11) (0.03) (0.85)
Mission Hill–Roxbury (part) −111.84 −71.60 0.00 −0.04 0.04 1.12
Dorchester (25.69) (33.71) (0.06) (0.14) (0.05) (1.39)
Mattapan–W. Roxbury −109.26 −91.23 0.10 0.13 0.05 2.88
Hyde Park (25.81) (33.54) (0.05) (0.10) (0.03) (0.66)
Brookline–Newton 197.93 207.45 0.09 0.02 0.00 −0.36
(35.53) (39.05) (0.04) (0.12) (0.02) (0.70)
Cambridge–Somerville 8.02 15.93 0.14 0.25 0.08 1.24
(23.94) (29.51) (0.05) (0.10) (0.02) (0.76)
Year Effects
1989 −18.48 156.15 −0.05 −0.36 −0.07 1.77
(87.09) (71.50) (0.14) (0.38) (0.03) (1.74)
1993 22.10 123.80 −0.20 −0.29 −0.20 2.27
(90.94) (123.03) (0.19) (0.22) (0.09) (1.55)
1998 119.73 150.37 −0.18 0.05 0.01 4.67
(61.67) (64.38) (0.24) (0.34) (0.06) (0.84)
N= 3541 3445 3145 3047 3543 3542
Number of clusters 84 84 84 84 84 84
Notes. All regressions include quadratics and cubics for number of bedrooms and other rooms, categorical age and
number of building units, age of building interactions with year, and controls for type of heat and presence of central
air as well as zone and year fixed effects. Column (1) also controls for whether various utilities are included in the rent.
Standard errors are corrected for clustering on the zone year level. All regressions are weighted by the inverse probability
that the housing unit is in the sample. Severe problems include pipe and plumbing failures, heating failures and electrical
problems. Chronic problems include holes in wall or floor, chipped or peeling paint, plaster damage, loose railings, etc.
Rent and housing costs are in 1998 dollars.

In contrast, column (5) considers the effect of decontrol on what might be thought of as
chronic aesthetic rather than functional maintenance items such as broken paint or plaster, holes
in the walls or floors, and loose railings. The dependent variable is a dummy variable indicating
whether one of these problems is present. The estimates demonstrate that ending rent control
leads to a significant reduction in these maintenance problems. A unit was almost 6 percentage
points less likely to experience such problems once its zone is decontrolled. Though rent control
does not seem to lead to catastrophic maintenance failures, it appears to reduce the maintenance
performed on rental units. As landlords can be fined for allowing water and heat failures, but not
for cracked paint, this result is not surprising.
The final column of Table 4 considers the effect of rent control on the length of time a renter
stays in a unit. To the extent that artificially low rents reduce the mobility of the population they
impose inefficiency. People who would otherwise move away decide to stay in a controlled unit
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 145

to keep the advantage of an artificially low rent. Column (8) shows that decontrol is associated
with a decrease of renter stays of 1.84 years. This is sizeable when compared to the mean renter
stay of 6 years in the sample.14
Although the results of Table 4 are evocative, they could be an artifact of sample composition.
As discussed previously, the end of rent control caused units to shift from owner to renter status.
Thus the zone level changes in 1998 might not be due to changes of unit quality or rent caused
by rent control, but due to higher quality units entering the renter sample. To deal with this
possible criticism, Table 5 presents the same regressions using a composition constant sample.

Table 5
Effects of rent decontrol on housing—composition constant sample
Dependent variable rent cost severe # probs chronic length
(1) (2) (3) (4) (5) (6)
Decontrol Effects
Ever controlled zone ∗ 1998 82.37 56.20 0.00 −0.08 −0.06 −2.51
(22.97) (25.93) (0.05) (0.08) (0.02) (0.60)
Zone Effects–Controlled Zones
Allston–Brighton–Roslindale 76.94 84.41 0.07 0.04 0.03 −0.33
Jamaica Plain (23.87) (26.90) (0.04) (0.10) (0.03) (0.80)
Back Bay–Fenway–Downtown 150.38 151.93 0.01 −0.01 0.00 0.44
North End–South End (22.10) (22.49) (0.05) (0.10) (0.03) (0.82)
Charlestown– E. Boston– −46.91 −45.32 −0.06 −0.23 −0.06 2.88
S. Boston–N. Dorchester (26.87) (27.05) (0.05) (0.11) (0.03) (0.61)
Mission Hill–Roxbury (part) −76.34 −60.00 0.00 −0.04 0.04 0.63
Dorchester (31.29) (32.82) (0.07) (0.14) (0.06) (1.22)
Mattapan–W. Roxbury −48.08 −38.61 0.10 0.11 0.05 2.61
Hyde Park (22.64) (26.58) (0.05) (0.10) (0.03) (0.61)
Brookline–Newton 65.57 86.80 0.02 −0.04 −0.03 −2.14
(36.34) (49.58) (0.07) (0.17) (0.04) (0.65)
Cambridge–Somerville 124.37 129.26 −0.01 −0.11 0.01 −1.60
(25.55) (27.33) (0.04) (0.09) (0.03) (0.57)
Year Effects
1989 214.18 221.62 −0.05 −0.29 −0.08 1.01
(35.75) (38.36) (0.13) (0.19) (0.03) (0.69)
1993 359.33 652.06 −0.20 −0.52 −0.06 1.78
(33.87) (142.42) (0.19) (0.43) (0.11) (2.00)
1998 549.34 559.77 0.08 −0.09 −0.03 4.89
(78.02) (77.19) (0.16) (0.24) (0.05) (0.94)
N= 3476 3388 3098 3000 3490 3490
Number of clusters 84 84 84 84 84 84
Notes. All regressions include quadratics and cubics for number of bedrooms and other rooms, categorical age and
number of building units, age of building interactions with year, and controls for type of heat and presence of central
air as well as zone and year fixed effects. Column (1) also controls for whether various utilities are included in the rent.
Standard errors are corrected for clustering on the zone year level. All regressions are weighted by the inverse probability
that the housing unit is in the sample. Severe problems include pipe and plumbing failures, heating failures and electrical
problems. Chronic problems include holes in wall or floor, chipped or peeling paint, plaster damage, loose railings, etc.
Rent and housing costs are in 1998 dollars.

14 Measuring length of tenancy may be subject to the same sort of bias as measuring unemployment spells (we only
observe uncompleted spells), thus this coefficient may overstate the true effect of rent control, though the effect is still
likely to be large (see Ault, Jackson abd Saba [4]).
146 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

Table 6
Effects of rent decontrol—some specification checks
Dependent variable rent cost chronic length rent cost chronic length
(1) (2) (3) (4) (5) (6) (7) (8)
Decontrol Effects
Boston ∗ 1998 75.03 44.70 −0.06 −1.73
(25.51) (29.41) (0.03) (0.67)
Non-Boston Controlled ∗ 1998 104.96 93.40 −0.06 −1.78
(31.41) (34.66) (0.02) (0.53)
Comparison of Real and Hypothetical Experiments
Ever controlled zone ∗ 1998 97.25 70.18 −0.09 −1.46
(27.30) (31.28) (0.03) (0.77)
Ever controlled zone ∗ 1993 7.89 2.48 −0.06 0.61
(23.00) (23.52) (0.04) (0.94)
Ever controlled zone ∗ 1989 24.64 20.14 −0.03 0.29
(19.89) (21.57) (0.03) (0.73)
N= 3541 3445 3543 3543 3541 3445 3543 3543
Number of clusters 84 84 84 84 84 84 84 84
Notes. All regressions include quadratics and cubics for number of bedrooms and other rooms, categorical age and
number of building units, age of building interactions with year, and controls for type of heat and presence of central
air as well as zone and year fixed effects. Columns (1) and (5) also control for whether various utilities are included in
the rent. Standard errors are corrected for clustering on the zone year level. All regressions are weighted by the inverse
probability that the housing unit is in the sample. Chronic problems include holes in wall or floor, chipped or peeling
paint, plaster damage, loose railings, etc. Rent and housing costs are in 1998 dollars.

This sample is obtained by omitting all the 1998 units that were owner occupied in 1990. This
new sample is composition constant because all of its units were rented during the rent control
regime as well as after control ended.
The results in Table 5 indicate that compositional issues are not a significant worry in inter-
preting the original evidence. The results are very similar with the exception of the renter stay
estimate, which is significantly larger (in absolute value) in the composition constant sample.
This makes sense. Only including apartments rented in 1990 excludes newer renters in converted
buildings from the sample.
Table 6 presents a series of specification checks on the Table 4 results. The first four columns
divide the decontrolled zones into Boston and other categories. The rent and cost effects are
somewhat smaller in the Boston zones while the maintenance and renter stay results are almost
the same across all zones. The final four columns add interaction terms to mimic hypothetical
policy changes conducted in 1993 and 1989 and compare them to the real end of rent control in
1998. In all cases the real policy change provides significant results while the imaginary experi-
ments result in small and insignificant coefficients.

4.3. Welfare

The preceding estimates allow a rough approximation of what the excess burden of Massa-
chusetts rent control regulations would have been if they were in force in 1998. If rent control
reduces the probability an apartment is available for rent by around 6 percentage points, as sug-
gested by the estimates of Tables 2–3, its continuation would have meant the loss of around
15,000 units from the Boston area rental market in 1998 relative to the number of rentals actually
observed. Additionally, the estimates of Table 4 suggest that if rent control had continued, the
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 147

counterfactual rent received by landlords would have been approximately 84 dollars below the
observed 1998 equilibrium rent of 741 dollars.
However, some of this rent differential accounts for the reduction in housing quality of rent
controlled units rather than reduction of the quantity of units supplied. Since the chronic damage
problems found in rent controlled housing would lead to a 20% rent reduction if that unit was in
a non-controlled zone, and an additional 6% of units in rent controlled zones will develop these
problems, I assume that rents would decrease about 1.2% or 9 dollars from the actual 1998 rent
due to a decline in unit quality. This leaves a counterfactual rent decrease of 75 dollars below
equilibrium level due to the price ceiling on rental housing.
To complete the deadweight loss calculation, I approximate the counterfactual price deman-
ders would be willing to pay for the reduced housing stock. For this calculation, I conservatively
assume that the price elasticity of demand for housing is −0.5, which implies demanders would
be willing to pay around 90 dollars more than the observed equilibrium price once rent control
lowers the available quantity of rental housing.15 Under these assumptions, the deadweight loss
from rent control is approximated by a triangle whose base is the 165 dollar wedge between the
price of demanders and suppliers and whose height is the 15,000 unit drop in quantity. This dead
weight loss of 1.2 million dollars a month represents about 29 dollars per actively controlled
apartment. This excess burden from underprovision of rental housing is in addition to the excess
burden from misallocation of rental units discussed by Glaeser and Luttmer [11].
Proper policy analysis compares the costs of such a program with the benefits it generates. The
potential benefits of rent control arise through the transfer of surplus from landlords to tenants,
but are only realized if social preferences are such that more weight is placed on the recipients of
this surplus than its original owners. A rough calculation, using the above assumptions indicates
that the rent control policies in these three cities would have transferred an average of 75 dollars
for each of the 226,000 counterfactual rental units in controlled areas or about 17 million dollars
per month in total from landlords to tenants.16
However, there is little evidence that rent control programs effectively transfer this surplus to
tenants society might wish to help, such as the poor or minority households.17 Table 7 provides

Table 7
Distribution of rent controlled units by income and race
A. Income
Income quartile 1 2 3 4
Percent of controlled units 0.26 0.41 0.19 0.11

B. Race
White Black Hispanic
Racial distribution in controlled units 0.81 0.08 0.04
Racial distribution in controlled cities 0.66 0.22 0.04
Notes. The reported numbers reflect the percentage of rent controlled tenants that fit in each characteristic category. The
racial distribution in controlled cities is calculated from the 1990 census STF-1.

15 Common estimates of price elasticity of housing demand run from −0.3 to −0.75.
16 Because the expectation of future transfers is capitalized into the sale price, this surplus is lost by the landlords at
the time rent control was passed, who are not necessarily current landlords. Additionally some of this surplus may be
recouped by landlords that require an illegal key money payment in return for obtaining a lease in a controlled unit.
17 Olsen [20] discusses papers that explore this issue.
148 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

some initial evidence that the recipients of rent control benefits in Massachusetts might not have
been those that society traditionally seeks to target in housing assistance programs. Only 26%
of rent controlled apartments were occupied by renters in the bottom quartile of the household
income distribution, while 30% of units were occupied by tenants in the top half of this distrib-
ution. Hispanic and African-American minority groups were similarly underrepresented in rent
controlled housing, comprising only 12% or the rent controlled population but one quarter of
the cities with rent control. This suggests that much of the transferred surplus may have been
received by wealthier households.

4.4. Spillover effects of rent control

The rent effects and surplus transfer estimated above seem quite large. If the end of rent con-
trol raises all decontrolled zone rents by an average of 84 dollars, and if rent control only affects
units under active control (around one-sixth of the units in the zone), then my estimates imply a
504 dollar average rent reduction for each controlled unit due to rent control. It is possible, how-
ever, that rent control also affects units that are not actively controlled. In the case of Boston, this
may operate through the passive rent appeal system discussed earlier. Alternatively, a spillover
effect from rent control may affect the price of nearby uncontrolled units. Though the underpro-
vision of housing due to rent control might raise rents in the uncontrolled sector, the reduced care
given to rent controlled units may make the zones with rent control less desirable for those living
in non-controlled housing. This spillover effect due to sub-optimal maintenance may decrease
all rents in an area.18
Estimates of Eq. (4), which permits rent control to affect both controlled and non-controlled
units, are presented in Table 8. These regressions contain the same unit characteristic controls
as earlier regressions as well as zone and time fixed effects, though these are not reported in the
table. Columns (1) and (2) show estimates of the effect of being a rent controlled unit on rent
with the potential spillover coefficient, λ constrained to equal zero.
The Least Squares estimates in panel A indicate that rent control is associated with a 170–180
dollar monthly decrease in rent levels. However, because rent controlled status is not randomly
assigned, this coefficient may not have a clear causal interpretation. Panel B addresses this
through Two-stage Least Squares Estimation of Eq. (3). In this model the instruments are in-
teractions between certain unit characteristics and whether the unit is in a zone that had rent
control, as well as the policy instrument. The characteristics include whether the unit is a single
family house, whether the unit is an owner occupied 2–3 family house, and whether the unit was
built before the rent control regulations were adopted. Each of these characteristics is used to
define eligible units for rent control, and hence is correlated with the rent control status of the
unit. The key identifying assumption is that the main effect of these characteristics on rent is the
same in the rent controlled zones as in the non-controlled zones.
The IV estimates are much larger than the corresponding Least Squares coefficients. They
indicate that rent control decreases the rent of a controlled unit by about 450 dollars a month,
and decreases housing costs by about 340 dollars a month. This represents a large proportion of
the 750 dollar average rent for decontrolled zones in 1998.
Columns (3) and (4) relax the constraint on λ, allowing the percentage of rent controlled units
in a zone to affect the rent of the non-controlled units. The Least Squares results suggest that the

18 Heffley [14] uses a spatial equilibrium model to predict a decrease in the price of housing services among uncontrolled
units in the presence of rent control.
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 149

Table 8
Own unit and spillover effects of rent control
Dependent variable no spillover effects with spillover effects
rent cost rent cost
(1) (2) (3) (4)
A. Least Squares
Unit is controlled −172.26 −182.47 −204.60 −205.96
(28.53) (26.86) (25.01) (26.30)
Percent controlled in zone − − −230.53 −165.94
(94.07) (100.23)
B. Instrumental Variables
Unit is controlled −457.56 −341.19 −327.37 −351.38
(137.10) (148.89) (185.60) (204.25)
Percent controlled in zone − − −238.91 19.04
(191.69) (213.28)
C. First-Stage
Dependent variable controlled % controlled
Decontrolled zone ∗ 1998 −0.11 −0.10
(0.04) (0.02)
Built before construction exemption ∗ Controlled zone 0.06 0.01
(0.03) (0.01)
Exemptible by owner occupied provision ∗ Controlled zone −0.03 0.007
(0.02) (0.004)
Exemptible as single family dwelling ∗ Controlled zone −0.21 −0.05
(0.04) (0.02)
F -value (4, 83 degrees of freedom) 14.17 11.45
N= 3544 3488 3544 3488
Number of clusters 84 84 84 84
Notes. All regressions include quadratics and cubics for number of bedrooms and other rooms, categorical age and
number of building units, age of building interactions with year, and controls for type of heat and presence of central
air as well as zone and year fixed effects. Columns (1) and (3) also controls for whether various utilities are included in
the rent. Standard errors are corrected for clustering on the zone year level. All regressions are weighted by the inverse
probability that the housing unit is in the sample. Rent and housing costs are in 1998 dollars. The reported F tests have
as null hypotheses that the coefficients on all instruments jointly equal zero.

effects of control on controlled units are sizeable, decreasing rent about 200 dollars a month. The
spillover effects are smaller but also statistically significant. The coefficients imply that having
10–12% rent controlled units in your zone will decrease your rent by 23–28 dollars a month.
As previously mentioned, a possible explanation for this result might be chronic maintenance
problems caused by rent control.
The IV estimates of this relationship are also suggestive. They indicate that rent control lowers
the rent of controlled units by about 320 dollars, while the estimated spillover coefficient is about
the same as under Least Squares. However, these estimates are insufficiently precise to conclude
that either coefficient is significant at anything smaller than a ten percent level. The coefficient on
the potential spillover effect for housing costs is actually positive but very imprecisely estimated.
Unfortunately, there are further barriers to a causal interpretation of these estimates. The first
is the possibility that the decreased quantity of rental housing in controlled zones will also have
spillover effects that increase the price in non-controlled zones. This would bias all my estimates
of the rent effect (e.g. from Table 4) as it violates a central assumption that rent control exerts
no effects on the non-controlled zones. However, work on estimating these cross area effects
150 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151

by Early and Phelps [9] suggests that their occurrence may be limited to the early years of a
rent control regime (which for Boston was the 1970s). This agrees with intuition. Since the non-
controlled areas in the Boston area are also those with fewer land use regulations and looser
permitting rules, they can adjust to the supply change in neighboring zones over time. Neverthe-
less, it is impossible using my empirical strategy to test for the presence of such effects.19
A second concern is the measurement of rent controlled status. The difference between ad-
ministrative estimates of the extent of rent control and the number of units reporting controlled
status in the survey suggest that tenants underreport rent control. Because the resulting measure-
ment error in both the controlled status and percentage controlled variables is non-classical the
use of instrumental variables will not correct any measurement error bias. The use of adminis-
trative estimates of the percentage of rent controlled apartments in a zone rather than an average
calculated from the sample results in only slight changes to the Table 8 estimates, suggesting that
the bias may not be a large problem. However, given the large standard errors and the impossi-
bility of ruling out measurement error bias the estimates of Table 8 are far from definitive, but
provide some suggestive evidence of a moderate spillover effect from rent control.

5. Conclusion

The sudden end of rent control in Massachusetts in 1995 provides a natural experiment to
study the effects of rent control. My results indicate that the intuition presented in simple micro-
economic models is correct. Rent control decreases the quantity of rental units supplied, as well
as rent and unit maintenance. It also lengthens renter stays. In addition, some evidence suggests
that rent control produces small spillover effects that decrease the rent of uncontrolled units in
controlled areas.
Further work could examine longer term impacts of ending rent control as well as investigate
the distribution of the surplus shifted by controls. If much of the benefit accrues to white upper
income households, rent control may prove to be an ineffective transfer program as well as an in-
efficient one. Researchers might also focus on investigation of spillover effects from rent control
policies and whether they are the result of lower maintenance levels in rent controlled areas.

Acknowledgments

I would like to thank Joshua Angrist, Daron Acemoglu, David Autor, Norma Coe, Alexis
Leon, Michael Steinberger, the editor, two anonymous referees and participants in the MIT and
BYU workshops for helpful comments and suggestions. Remaining errors are the author’s sole
responsibility.

References

[1] D. Acemoglu, J. Angrist, How large are human capital externalities? Evidence from compulsory schooling laws,
NBER Macro Annual 15 (2000) 9–59.
[2] R.M. Alston, J.R. Kearl, M.B. Vaughan, Is there a consensus among economists in the 1990s? American Economic
Review 82 (1992) 203–209.

19 It might be possible to set some bounds on the rent effect using the elasticity of substitution in the demand for housing
between zones Unfortunately, there is almost no empirical evidence of what the true elasticity of substitution might be
for the Boston Area.
 D.P. Sims / Journal of Urban Economics 61 (2007) 129–151 151

[3] R. Arnott, Time for revisionism on rent control, Journal of Economic Perspectives 9 (1995) 99–120.
[4] R.W. Ault, J.D. Jackson, R.P. Saba, The effect of long-term rent control on tenant mobility, Journal of Urban
Economics 35 (1994) 140–158.
[5] Boston Bar Association, Rent, Eviction, and Condominium Controls in Boston, Boston Bar Association, Boston,
1987.
[6] Brookline Town, Town of Brookline Rent Control By-laws and Rules, Brookline Town, Brookline, 1991.
[7] Cambridge Rent Control Board, The Cambridge Rent Control Board Enabling Act, Ordinance, Regulations, Etc.,
Cambridge Rent Control Board, Cambridge, 1993.
[8] D.W. Early, Rent control, rental housing supply, and the distribution of tenant benefits, Journal of Urban Eco-
nomics 48 (2000) 185–204.
[9] D.W. Early, J.T. Phelps, Rent regulations’ pricing effect in the uncontrolled sector: An empirical investigation,
Journal of Housing Research 10 (1999) 267–285.
[10] W.F. Galvin, What you should know if you are a tenant or landlord of a rent controlled unit in Boston, Brookline or
Cambridge, Society of the Commonwealth Citizen Information Service, Boston, 1995.
[11] E.L. Glaeser, E.F.P. Luttmer, The misallocation of housing under rent control, American Economic Review 93
(1999) 1027–1046.
[12] J. Gyourko, P. Linneman, Equity and efficiency aspects of rent control: An empirical study of New York City,
Journal of Urban Economics 26 (1989) 54–74.
[13] J. Gyourko, P. Linneman, Rent controls and rental housing quality: A note on the effects of New York City’s old
controls, Journal of Urban Economics 27 (1990) 399–409.
[14] D. Heffley, Landlords, tenants and the public sector in a spatial equilibrium model of rent control, Regional Science
and Urban Economics 28 (1998) 745–772.
[15] N.K. Kutty, The impact of rent control on housing maintenance, Journal of Housing Studies 11 (1996) 69–88.
[16] C. Moon, J.G. Stotsky, The effect of rent control on housing quality change: A longitudinal analysis, Journal of
Political Economy 101 (1993) 1114–1148.
[17] B. Moulton, An illustration of the pitfalls in estimating the effect of aggregate variables on micro units, Review of
Economics and Statistics (1990) 334–338.
[18] E.O. Olsen, An econometric analysis of rent control, Journal of Political Economy 80 (1972) 1081–1100.
[19] E.O. Olsen, What do economists know about the effect of rent control on housing maintenance? Journal of Real
Estate Finance and Economics 1 (1988) 295–307.
[20] E.O. Olsen, Economics of rent control, Regional Science and Urban Economics 28 (1988) 673–678.
[21] US Census Bureau, American Housing Survey, Metropolitan Sample—Boston—Public Use Files with Documen-
tation, US Department of Housing and Urban Development, 1985, 1989, 1993, 1998.
[22] US Department of Housing and Urban Development, Building Permits Database, US Department of Housing and
Urban Development, 1980–2000.
[23] US Department of Housing and Urban Development, Report to Congress on Rent Control, US Department of
Housing and Urban Development, 1991.
[24] D. Vandenbroucke, Summary of Research into the 1998 American Housing Survey Metropolitan Sample Data
Anomalies, US Department of Housing and Urban Development, 2000.
[25] J.M. Wooldridge, Cluster-Sample methods in applied econometrics, American Economic Review 93 (2003) 133–
138.