- 1 -

Christopher K. Hsee
J ean-Pierre Dub
Yan Zhang

*Christopher K. Hsee is Theodore O. Yntema professor of behavioral science and marketing at
the Graduate School of Business, University of Chicago, 5807 South Woodlawn Ave., Chicago
IL 60637, Email: chris.hsee@ChicagoGSB.edu. J ean-Pierre Dub is a professor of marketing
and Neubauer Faculty Fellow at the Graduate School of Business, University of Chicago, 5807
South Woodlawn Ave., Chicago, IL 60637, Email: jdube@ChicagoGSB.edu. Yan Zhang is a
Ph.D candidate at the Graduate School of Business, University of Chicago, 5807 South
Woodlawn Ave., Chicago IL 60637, Email: yzhang6@ChicagoGSB.edu.
- 2 -
A Behavioral Analysis of Shanghai Real Estate Prices
Abstract
In a field study, we identified an intriguing inconsistency between real estate developers desired
sales pattern (selling all apartments at the same rate) and the actual sales pattern (selling good
apartments faster). We explained this inconsistency with Tversky, Sattath and Slovic (1988)s
prominence principle, according to which buyers, who were in a choice mode, weighed the
desirability of floors more heavily than developers, who were in a matching mode when setting
prices. We corroborated our explanation with controlled experiments using potential Shanghai
homebuyers and actual Shanghai real estate developers as participants. To the best of our
knowledge, this is the first research to relate the prominence principle to a real-world issue in
one of the worlds most active markets.

- 3 -
A Behavioral Analysis of Shanghai Real Estate Prices
Since the new millennium, China has emerged as one of the worlds most formidable
consumer markets. In 2004, the size of the consumer market was estimated at 36 million urban
Chinese and, every year, roughly 20 million Chinese turn 18 (Grant, 2006). The impact of this
consumer boom has been particularly apparent in the Chinese real estate market, one of the most
active real estate markets in the world. During the late 1990s, the Chinese government enacted
several radical changes in the regulation of state-provided housing to stimulate economic
growth. In 1999, the state lifted a ban on the resale of privatized housing. These measures led to
an explosion in the Chinese housing market, which has subsequently been estimated to have
been contributeding 1.5% of annual GDP growth (Economist 2001). Rapidly rising housing
pricesin Shanghai, prices have nearly doubled between 2000 and 2004 (Economist 2005)
have fueled billions of dollars in housing construction. Since 1996, 67,333 residential properties
were sold in Shanghai. In 2000 alone, about 7.5 million square meters (80.73 million square feet)
of existing houses were sold in Shanghai, with a total transaction value of roughly RMB 65.6
billion (about U.S. $8 billion) (Ye, 2004). At a more micro level, real estate developers are
increasingly relying on careful pricing practices to avoid missing profitable opportunities in this
fast-paced, but otherwise relatively young housing market.
We begin with a field study that involves real estate developers in Shanghai. From the
study, we learn that most developers strategically set their prices to obtain a sales pattern
whereby all their units sell at the same rate over time. Yet despite their best efforts to price
apartments commensurate to their perceived quality by consumers, in practice sellers routinely
find themselves stuck with the less-desirable units selling at a much slower pace. Specifically,
apartments on the lowest floors, the least desirable floors in the Chinese market, are routinely
- 4 -
slower to sell. We confirm this good-faster-than-bad pattern empirically by looking at actual
price and sales data for specific Shanghai apartment complexes.
It should be noted at the onset that in this article we do not make any claims as to which
sales pattern is normatively optimal. The answer to this normative question would depend on
many additional factors including developers budgetary constraints and their rate of time
preference. Instead we are only interested in the discrepancy between developers desired sales
pattern and the actual sales pattern.
We explain the discrepancy between the desired and the actual sales patterns using
Tversky, Sattath, and Slovic (1988)s prominence principle. According to that principle, when
individuals choose among alternatives with no clearly dominant choice (i.e. an alternative is
superior on one attribute but inferior on another attribute), the option that is superior on the more
prominent attribute is more likely to be preferred in choice even though the two options are
equally attractive in matching. An attribute is defined as more prominent than another attribute if
laypeople perceive it to be more important than the other attribute, regardless of whether it is
more important in the regression sense. For example, most people would consider life as more
important than money, even though in the regression sense whether life or money is more
important depends on the ranges and the distributions of the two attributes.
In the context of the real estate development market, developers are in a matching task
when they set prices. Namely, they first set a base price for an average-quality unit in a building
and then make adjustments to set the prices for the other units so that all the units would are be
equally attractive to homebuyers. On the other hand, homebuyers are in a choice task, and as our
data show, they perceive floor to be more important (prominent) than price. Failure by
developers to account for the relative prominence between floor and price in the buyers choice
- 5 -
process may lead the former to under-price the more desirable floors and/or over-price the less
desirable floors.
To show that the inconsistency between developers intentions and buyers choices we
found in the field study was indeed due to the prominence principle, we conducted several
additional studies. First, we ran a survey designed (a) to show that the floors we assumed to be
undesirable in the field study were indeed considered undesirable by Shanghai homebuyers; and
(b) to show that floors were indeed perceived to be more important than prices in home-purchase
decisions.
We then conducted two controlled experiments to replicate the inconsistency found in the
field data. The first experiment consisted of two between-subjects conditionsone choice and
one matchingfor a pool of participants from Shanghai who had revealed the income potential
to buy a new apartment. In the matching condition, participants were asked to state prices that
would equalize their perceived values between two hypothetical apartments on different floors.
In the choice condition, we used the same two apartments and used the equalized prices from
the matching condition. Participants were then asked to choose one of the apartments. We
observed a striking inconsistency between price-setters and choosers. Participants in the choice
condition overwhelmingly chose the apartment on the better floor, despite the use of prices that
were revealed to equalize the attractiveness of the apartments, on average, in the matching
condition. In the second controlled experiment, we asked actual developers to set prices. Again,
we replicated the inconsistency between price setters and choosers. The results of these
controlled experiments were remarkably parallel to the results of the field study, thus reinforcing
our belief that the prominence principle was the underlying reason.
- 6 -
Previous research has documented laboratory evidence for the prominence effect, namely,
inconsistencies between matchers and choosers. To the best of our knowledge, we are the first to
make a connection between this behavior and real outcomes in the market place. Our findings
also shed light on a peculiar pricing anomaly in one of the worlds most turbulent and rapidly-
growing marketsthe Shanghai real estate market.
The remainder of the paper is organized as follows. In the first study, we summarize the
interviews of Shanghai developers. In the second study, we examine actual field sales data for
representative Shanghai developments. In the third study, we report a survey of potential
Shanghai home-buyers to show their floor preferences. We then discuss related literature in
judgment and decision-making to suggest an explanation for our stylized fact. In the fourth and
fifth studies, we report two controlled experiments to replicate the inconsistency observed in the
field. Finally, we conclude with a discussion of possible remedies.

Study 1: Interview of Realtors Regarding Their Desired Sales Pattern
The main purpose of this study was to examine developers desired sales patterns. In
addition, the study also explored why developers wanted the sales pattern as they indicated, and
how they set prices to achieve their desired sales pattern.

Method
We conducted extensive interviews with 47 experienced price-setters who closely
collaborated with an online website doing business in apartment trading and financial mortgage.
Their qualifications ranged from general managers of real estate companies, marketing managers
- 7 -
responsible for real estate pricing, to employees who had been working for more than two years
in the real estate industry with significant pricing experience.
In the interview, we first explained that sales pattern, by our definition, meant the sales
rate of different floors. Then we explained how price differences shaped sales patterns.
Specifically, they were told that if the price difference between good floors and bad floors was
relatively small, then good floors would sell faster than bad floors. If the price difference was
sufficiently large, good floors would sell slower than bad floors. Finally, there existed some
intermediate price difference such that good floors and bad floors would sell roughly at the same
rate. These three scenarios corresponded to three sales patterns: good-faster-than-bad pattern,
bad-faster-than-good pattern, and flat pattern.
We then asked interviewees which of these three sales patterns they attempted to achieve
when setting prices for apartments in a new apartment development. We also asked interviewees
to explain the reasons for their preference for the specific sales pattern they indicated. Finally,
we asked them for the pricing methods they used to achieve their desired sales pattern.

Results and Discussion
Based on the interviews, we learned that developers generally preferred either a flat or a
bad-faster-than-good sales pattern in most cases. Of the 47 interviewees, 26 wanted the flat
pattern, 16 wanted the bad-faster-than-good pattern, 3 wanted the good-faster-than-bad
pattern, and 2 did not give clear answers. The interview responses provided overwhelming
support (89% of interviewees) for a sales pattern in which bad apartments sell at least as fast
as good apartments.
- 8 -
Developers also provided several explanations for why they would want apartments on
less desirable floors to sell at least at the same rate as good apartments. A very common
explanation was that good apartments were important for attracting potential consumers who
need to expend time and effort to visit a development in the first place. If most of the units on
good floors had been sold out, a building had more difficulty attracting potential consumers.
Some developers speculated that if units on good floors were unavailable, consumers would not
even bother to come to visit a new building. In general, if there were only bad floors available,
developers would have to resort to other means of attracting consumers, such as aggressive price
reductions that often lowered the developers profit.
Developers also provided cultural reasons why they would prefer not to sell the
apartments on the best floors first. In China, it was common for people to live close to their
relatives (including parents and children) and friends. Therefore, friends and relatives often
bought apartments in the same building. For social reasons, individuals would not want to
purchase apartments on inferior floors. A developer would prefer to have early purchasers buy
apartments on relatively undesirable floors because it would be much easier subsequently to sell
to friends and family on more desirable floors. In contrast, if early buyers purchased on the better
floors, it was much more difficult to sell units subsequently to their friends and family on a less
desirable floor. For these social reasons, developers preferred to sell the less desirable floors
first.
The interviews also provided accounts of the pricing strategy used to achieve the desired
sales pattern. To determine the list price, developers first chose an average quality unit in a
building. By dividing the total desired profit from the building by the total square meters, they
obtained the base price per square meter for this average unit (i.e. they used cost-plus pricing).
- 9 -
They adjusted the prices of the remaining units up or down depending on their quality relative to
the average unit. For the floor of an apartment, they typically used a constant price adjustment,
assuming that higher floors were more desirable. For instance, they might pick an average
quality floor such as the 5
th
floor and set the base price at 6000 RMB per square meter. They
would then add 300 RMB per square meter for every floor above the 5
th
, and deduct 300 RMB
per square meter for every floor below the 5
th
. In addition to assigning an upward price
adjustment to higher floors, developers might make further adjustments, e.g., assign a downward
adjustment to the price of the top floor, as it may get very hot during the summer. Floor was not
the only price differentiator. Developers might also make price adjustments based on other
factors such as exposure and floor plan.
To determine the price adjustment between floors in a building, developers typically took
the following steps. First, they surveyed potential consumers with questions such as: if the 5
th

floor is worth 6000 RMB per square meter, then how much do you think the 6
th
floor is worth?
These types of stated-preference data generated preliminary estimates of consumer willingness-
to-pay for floors. To gauge the external validity of these estimates, some developers visited
buildings in the surrounding area to measure the competitive prices. Developers might also use
crude checks, such as comparing results with their own personal views of what they would be
willing to pay for a 6
th
floor apartment given the availability of a 6000 RMB 5
th
floor unit. In
general, however, the price adjustment for floors was obtained via a matching strategy whereby
floors were priced according to their perceived value relative to a base level.
Based on the interviews, we provided quantitative support for the selling strategy of
experienced developers who preferred to sell units on less desirable floors at least as fast as those
on good floors. We also provided anecdotal accounts of the pricing strategy used to obtain this
- 10 -
outcome. Finally, we provided anecdotal accounts of the actual sales patterns of apartments,
which were inconsistent with the intended patterns. In the next section, we examine actual field
sales data to provide quantitative support for the list pricing practices and the actual sales
patterns described in the interviews.

Study 2: Field Sales Data
Study 1 showed that most developers intended to sell all units in a building at the same
rate. We now examine actual sales patterns of several representative buildings in Shanghai to
verify whether they are consistent with the desired sales pattern.

Method
We selectively used developments from three major residential areas in Shanghai to obtain
a roughly representative account of the real estate market. The three properties contained both
mid-rises and high-rises targeted to consumer groups ranging from middle-income employees to
relatively rich individuals. In each of the properties studied, all units were listed on the market
from their launch dates. Hence, developers of these properties did not tactically withhold good
apartments to manipulate the sales rate.

Results and Discussion
Dataset 1: Four 6- or 7- floor Midrises
Our first dataset came from a gated community consisting of two six-story and two seven-
story mid-rises. Each floor in each building had two units with an average unit size of 140 square
meters. All the units faced south. The four buildings were initially offered in November 1999. In
- 11 -
Figure 1, we report the average price per square meter for each floor across these four buildings.
Consistent with our interviews with developers, prices per square meter increased with the floor
on which an apartment was situated, with the exception being the top floor.
To assess the sales pattern across the four buildings, we report the average number of
months for a unit to sell by floor in Figure 2. We will provide a quantitative analysis of the sales
pattern in the next section. Here, we just want to make a qualitative observation. Contrary to the
intentions of most developers, apartments on different floors seemed to sell at different rates. For
example, units on the 4
th
floor took only 15.1 days to clear the market, but units on the 1
st
floor
took 20.8 days.

Dataset 2: Seven 11-floor Midrises
Our second dataset came from seven mid-rises located in a newly-developed
neighborhood in Shanghai. Each building had 11 floors, with two units on each floor; the
average size of each unit was 150 square meters. All seven buildings were initially offered in
September 2001. In Figure 3, we report the average list price for an apartment by floor.
Again, contrary to most developers intentions, units on different floors sold at different
speeds. In Figure 4, we report the average number of months an apartment was on the market by
floor. The sales duration for apartments in the seven residential buildings again seemed quite
different.

Dataset 3: A 30-floor Highrise
In the two previous field datasets, we examined the sales pattern of mid-rises. The third
dataset corresponded to a thirty-story high-rise building with twenty units on each floor. With
- 12 -
the exception of floors 28 and 39 which had larger units, each floor had the same floor plan.
Floors 1 to 3 were for commercial use and not for sale. The unit sizes ranged from 25 square
meters to 105 square meters. Figure 5 shows the average list price for each floor.
Figure 6 shows the duration in months that an apartment was on the market since initial
offering. Again, we observe systematically different sales speeds for units on different floors.
In summary, all three datasets suggested that contrary to the intentions of most developers,
units on different floors did not sell at the same speed. Furthermore, the sales pattern across
floors did not seem random: On average, apartments on low floors appeared to sell more slowly
than apartments on other floors.

Study 3: Survey of Potential Shanghai Apartment Buyers for Their Attitudes about Floors
The graphs in the previous section suggested that units on different floors sold at
systematically different speeds: slow-selling units were generally on low floors and fast-selling
units were generally on middle or high floors.
In China, particularly in Shanghai, low floors are considered undesirable because units on
low floors usually are damp, have poor views, and are susceptible to mosquitoes. Thus, we
hypothesize that slow-selling units are generally on undesirable floors and fast-selling units are
generally on desirable floors.
To test this hypothesis, we need a more fine-tuned definition of good and bad floors.
Instead of defining good and bad floors ourselves, Study 3 asked potential homebuyers to tell us
which floors were good and which floors were bad. We then treated their judgments as the
operational definition of good and bad floors and reanalyzed the three datasets in Study 2.

- 13 -
Method
We conducted a survey of 84 participants in a short-term non-degree executive program in
a large university in Shanghai. The ages of the participants ranged from 28 to 55, with most of
the participants in their mid thirties. Most (93.5%) of the participants had purchased homes in
Shanghai.
In the questionnaire, they indicated whether floor or price was more important to them
when considering buying an apartment in Shanghai. They were then asked to consider buying an
apartment in a hypothetical 7-story building with identical floor plans for each apartment. For
this building, participants were asked to indicate which floor(s) they would consider as bad
floor(s). Similar exercises were then carried-out in the context of a hypothetical 11-story
building and then a 30-story building for which the first 3 floors were not for sale. We chose
these three hypothetical buildings to mimic the three types of buildings in our field study.

Results and Discussion
The results from the survey confirmed the strong role of floor. Seventy two percent of the
respondents indicated that floor was more important than price.
The survey also clarified which floors buyers considered bad. We define a floor as bad if more
than 50% of respondents judged it bad. In the latter part of this section, we will use this
characterization of good versus bad floors to reanalyze the field datasets we presented in study 2
and test whether bad floors sell significantly slower than good floors.
In the 7-story building, the 1
st
and the 2
nd
floors were considered bad. In the 11-story
building, the 1
st
and the 2
nd
floor were considered bad. In the 30-story building, the 4
th
, the 5
th
,
and the 30
th
floors were considered bad.
- 14 -
Interestingly, the relative position of a floor in a building seemed to dictate whether it was
perceived as bad or good as opposed to the absolute floor number. For example, in the 30-
story building, the relatively low 4
th
and 5
th
floors were considered bad even though this was not
the case in the smaller 7 and 11 story buildings. In the 30-story building, the highest floor was
also considered a bad floor.

Analysis of the Field Sales Data Based on Buyers Attitudes about Floors
We now revisit the field sales data (Study 2). Recall from Figures 2, 4 and 6 that we
observed different selling rates for different floors. Using the survey (Study 3) results to define
good and bad floors, we could empirically test whether bad floors in the field study (Study 2)
indeed sold more slowly than good floors.
The results supported our hypothesis that units on bad floors sell slower than units on
good floors. In Dataset 1 the bad floors stayed on the market for 21.00 months whereas the good
floors for only 15.00 months (t=2.19, p<0.05). In Dataset 2 the bad floors stayed on the market
for 36.32 days whereas the good floors for only 26.10 days (t =2.77, p<0.01). In Dataset 3 the
bad floors stayed on the market for 10.31 months whereas the good floors stayed for only 6.40
months (t=5.45, p<0.0001).
These results provided quantitative support for the apparent empirical puzzle: the actual
sales patterns differed from the pattern intended by sellers. Given sellers intentions, the patterns
in prices and sales suggest that sellers were systematically under-pricing good floors (or over-
pricing bad floors).

- 15 -
Theory: Pricing versus Choice
We have now confirmed that consumers indeed considered the floor on which an
apartment was situated to be more important than its price. Apparently, developers were aware
of the importance of floor. They attempted to gauge consumer willingness-to-pay for floors
when adjusting prices of units on different floors. The observed inconsistency between the actual
sales pattern and the developers desired sales pattern seems perplexing. Developers appeared to
be systematically failing to set the relative prices of apartments on good and bad floors to
achieve the desired sales pattern. We now describe theory that could potentially explain the
underlying source of this bias in pricing.
When homebuyers decide which unit to purchase, they are in a choice mode: they make
comparisons between units and choose the one they like most. In the choice mode, a homebuyer
selects a unit from an offered set of two or more units after comparing different attributes such as
price and floor. However, when developers set apartment prices, they are in a matching mode:
they anchor on a base price and then adjust the prices to make all the units equally attractive to
consumers. Normatively speaking, consumers should be indifferent between apartments that,
through matching, have been priced to appear equally attractive.
However, established research in judgment and decision-making demonstrates that people
do not have well-defined preferences (e.g., Liu & Soman, 2006; Payne, Bettman and Schkade,
1999; Slovic, 1995). Different preference elicitation procedures may highlight different aspects
of options, suggest alternative judgment strategies, and reverse or alter preferences.
A classic example of preference reversal is the p-bet/$-bet reversal offered by
Lichtenstein and Slovic (1971). Consider the following pair of lotteries:
A: .95 Win $2.50; .05 Lose $.75
- 16 -
B: .40 Win $8.50; .60 Lose $1.50.
Lottery A (called a P-bet) gives a large probability of winning a small amount of money;
lottery B (called a $-bet) gives a smaller probability of winning a larger amount. Participants
were either asked to pick the lottery they wanted to play (the choice condition) or to price the
two lotteries (the pricing condition). Normatively, these two elicitation methods should yield the
same preference order. But Lichtenstein and Slovic documented a robust preference-reversal. In
the choice condition, most participants chose the P-bet, but in the pricing condition, most
assigned a higher price to the $-bet. The choice-pricing preference-reversal has been replicated
by many other researchers (e.g., Grether & Plott, 1979; Pommerehne, Schneider & Zweifel,
1982; Reilly, 1982), and also replicated in casinos with real gamblers and real money (e.g.,
Lichtenstein and Slovic, 1973).
Extending the original p-bet/$-bet preference reversal, Slovic (1975) and Tversky, Sattath
and Slovic (1988) identified another robust reversal. Participants were presented with
information about two hypothetical job candidates applying for a production engineer position
who differed on two attributes, technical knowledge and human relations, as follows:
Technical Knowledge Human relations
Candidate A 86 76
Candidate B 78 91
Both attributes were rated on a 40 (very weak) to 100 (superb) scale. The participants were
told that technical knowledge was more important than human relations. One group of
participants (the choice condition) was asked to choose between the two candidates. Another
group of participants (the matching condition) was presented with the same two alternatives,
with one of the four scores missing, and was asked to fill in that missing score so that the two
- 17 -
candidates were equally attractive. In choice, most people chose Candidate A (the one with the
better technical score), but in matching, the score most people filled in suggested that they would
have preferred Candidate B if that score had not been missing and had been the same as
presented in the choice condition. This finding suggested a prominence effect -- that the more
important attribute in a choice set (e.g., technical knowledge in the study) receives more weight
in choice than in matching. This finding has been replicated in various forms by other
researchers (e.g., Auh & J ohnson, 2005; Carman & Simonson, 1998; Chernev, 2005; Fischer &
Hawkins, 1993; Nowlis & Simonson, 1997).
Tversky et al. (1988) proposed the compatibility principle to explain the choice-matching
reversal (see also Slovic, Griffin, & Tversky, 1990). The weight of a stimulus attribute is
enhanced by its compatibility with the task. Generally speaking, qualitative information about
the ordering of the dimensions weighs more in the ordinal method of choice than in the cardinal
method of matching; whereas quantitative information weighs more in matching than in choice.
Our survey of potential apartment buyers indicated that the floor on which an apartment was
situated figured prominently in consumer preferences. Given the compatibility rule, in choice
mode we expected the floor of an apartment, the qualitative attribute, to receive more weight
than in matching mode. When developers used matching to determine the price adjustment for
apartments on different floors of a building, they assigned relatively less weight to floor. In
essence, the developers use of matching led to a bias in their measure of consumer willingness-
to-pay for floor because the actual demand for apartments was generated by choice.
In the next section, we present two controlled studies to replicate the inconsistency
between price-setters and buyers, as we observed in the field data.

- 18 -
Study 4: Consumer Pricing versus Consumer Choice
To make a connection between the experiments and the Shanghai real estate market, we
used potential home-buyers from Shanghai as participants. In the pricing condition, we asked
potential consumers to set the prices for two units in a hypothetical building with the objective of
making the two units equally attractive to them. We then used the prices generated from the
participants in the pricing condition to set the prices in the choice condition. In the choice
condition, we used the same two hypothetical units with prices set at the average levels from the
pricing condition. A separate group of participants was asked to choose among the two units.

Method
We conducted this study using 140 respondents who were either students in a large
business school in Shanghai or employees in two companies in Shanghai who had worked in the
city for three to five years.
Participants were assigned to either a pricing or a choice condition. In both conditions,
participants were asked to imagine that after graduation, they planned to buy a two-bedroom
unit. They were told that they were interested in a newly-built, 15-story building in which only
two units were left. The two units had the same 100 square meter floor plan. The only difference
was their floor location: Unit 101 was on the 1
st
floor and Unit 801 on the 8
th
floor.
In the pricing condition, participants completed two tasks in which they were told the
price of one unit and were asked to state the price for the other unit that would make the two
units equally attractive to them. In the first task, they were told that Unit 101, on the first floor,
had a price of 500,000 RMB and then asked to state the indifference price for Unit 801, an
identical unit on the eighth floor. The second task was identical except that they were told the
- 19 -
price of Unit 801 was 500,000 RMB and were asked to price Unit 101. The order of Unit 101
and Unit 801 was counter-balanced.
In the choice condition, participants were again asked to perform two tasks. In this case,
they were asked to make a choice between units 101 and 801. To be conservative, we designed
the choice condition based on the matched prices from the pricing condition to make the choices
as similar as possible on average. Hence, in the first task, participants were asked to choose
between Unit 101 priced at 500,000 RMB and Unit 801 priced at its average stated price from
the pricing condition, where we rounded the price up to the next highest integer value to make it
realistic. In the second task, participants were asked to choose between Unit 101 priced at the
average stated price from the pricing condition and Unit 801 priced at 500,000 RMB. Again, we
counter-balanced the ordering of these two tasks.

Results and Discussion
In the choice condition, we predicted that the prominent attribute, floor, would weigh
more heavily than in the pricing condition. This prominence would make the unit on the 8
th
floor
more attractive to consumers in the choice condition. Thus, we expected to observe a pricing-
choice preference reversal whereby, in the choice condition, the unit on the 8
th
floor would have
been more frequently chosen than the unit on the 1
st
floor despite their equivalence to consumers
in the pricing condition.
To test for a reversal between the choice and pricing conditions, we proceeded as follows.
Under choice, we simply looked at the choice frequencies of the two units. Under pricing, we
used the matched prices to infer what the participants choice would have been had they been in
the choice condition. For instance, under one of the choice tasks, we presented Unit 101 at
- 20 -
500,000 RMB and Unit 801 at 560,000 RMB. We inferred that any participant in the pricing
condition who stated a matched price for Unit 801 above 560,000 RMB when Unit 101 was
500,000 RMB would have chosen the former. In the second choice task, Unit 101 was priced at
440,000 RMB and Unit 801 was priced at 500,000 RMB. We inferred that anyone in the pricing
condition who stated a matched price for Unit 101 above 440,000 RMB when Unit 801 was
500,000 RMB would have chosen the former.
As expected, floor played a more prominent role in the choice condition than in the pricing
condition. Under choice, participants presented with an offer of Unit 101 at 500,000 RMB and
Unit 801 at 560,000 RMB chose the latter 85.1% of the time. Participants in the pricing
condition were inferred to choose Unit 801 only 30.3% of the time. Similarly, under choice,
participants presented with an offer of Unit 101 at 4400,000 RMB and Unit 801 at 500,000 RMB
chose the latter 78.4% of the time. Participants in the pricing condition were inferred to choose
Unit 801 only 33.3% of the time (85.1% versus 30.3%, two-way chi-squared =43.45, p<0.0001,
and 78.4% versus 33.3%, two-way chi-squared =28.90, p<0.0001, respectively).
Based on the stark difference in choice probabilities under pricing versus choice, we
expect that Unit 801 would sell faster than Unit 101 if both units were put on the market
simultaneously and priced using the matching policy described by price-setters in Study 1.
Moreover, this sales pattern would mimic the one observed in the field sales data in Study 2
capturing the inconsistency between price setters desired sales pattern and the patterns observed
empirically in the market place.

Study 5: Developer Pricing versus Consumer Choice
Study 5 extended Study 4 in two important directions. First, respondents considered all the
- 21 -
units in a hypothetical building instead of only two units. Second, the unit prices in the choice
task were set by professional developers rather than consumers.

Method
As in study 4, study 5 had two between-subjects conditions: one pricing condition in
which developers were asked to set prices for the units in a hypothetical building and one choice
condition where consumers were asked to rank the units according to their preference orders.
Unlike study 4, which used only two apartments, study 5 used an entire building with 10 units to
make the price-setting process more realistic for the developers who participated in the study.
In the pricing condition, the respondents were the same 47 developers surveyed earlier in
our field study, who generally desired a flat or a bad-before-good sales pattern. Their task was to
set unit prices for a hypothetical building before the building was listed on the market. They
were asked to imagine that the building was located in a nice residential neighborhood in
downtown Shanghai. The building had 10 floors and each floor had one 100 square meter unit.
Each unit had a living room and two bedrooms, with the living room and one bedroom facing
south, and the other bedroom facing north. The average per-square meter cost of the building
was 5,000 RMB and, thus, the average per-apartment cost was 500,000 RMB.
In the choice condition, the respondents consisted of 43 potential homebuyers recruited
from a population of participants in a non-degree executive program in a large university in
Shanghai. Virtually all of them either already owned an apartment or could afford one in
Shanghai. These respondents were told that they were going to buy a unit in a newly developed
building in which they were interested. The description of the hypothetical building was the
same as in the pricing condition, except that the prices of each unit, instead of the costs, were
- 22 -
presented to the participants. We set the price of each apartment in the choice condition at the
average recommended price across the developers. As in study 4, we felt the use of the average
prices from the pricing condition would homogenize the relative attractiveness of apartments to
some extent, making the design more conservative. Participants were first asked to assume that
all the units were available and to choose the unit they liked the most. They were then asked to
assume that the unit they liked the most was no longer available and to choose the one they liked
the most among the remaining units. They repeated this procedure until all the floors were
exhausted. This method provided us with each consumers ranking scheme for the 10 apartments.
The procedure also mimicked what typically happens during the sales of a building in real life:
when more desirable units are sold, potential buyers could only choose among the remaining
units.

Results and Discussion
The results not only replicated the findings of Study 4 but also closely resembled the
findings of the field study.
In Figure 7, we report the prices set by the developers. As expected, we observed an
almost linear upward adjustment for units on successively higher floors, with a slight downward
adjustment for the top floor. This pattern was consistent with the price adjustment process
described by developers in study 1. Furthermore, the relatively low pricing on lower floors and
the downward adjustment on the top floor were consistent with our findings in studies 1 and 3
whereby both consumers and developers associated these floors as less desirable.
As in study 4, we again expected floor to weigh more heavily in the choice condition than
in the pricing condition. To construct the test, we used the results from the 11-story building in
- 23 -
study 3 to define the first two floors as bad, and the remaining floors as good. A participant was
deemed to have chosen a good floor (bad floor) if her top-ranked floor was a good floor (bad
floor). Under choice, we simply used the observed ranks. Under pricing, we used each
developers reported price-differential between each pair of floors to determine her
corresponding floor price premium. Hence, if a developer set a 500 RMB price difference
between the 3
rd
and the 6
th
floors, we concluded that she perceived 500 RMB to be the monetary
equivalent of the difference in value between those two floors (i.e. 500 RMB was interpreted as
the compensating differential for the difference in floor). We then used these differentials to infer
each developers hypothetical ranking scheme had they been in the choice condition. For
instance, under choice, the unit on the 3
rd
floor was 6280 RMB and the unit on the 6
th
floor was
6780 RMB. For any developer who had set a price-differential between the 3
rd
and 6
th
floors of at
least 500 RMB, we would infer a higher ranking for the 6
th
floor. For the few instances where a
developer would have been indifferent (i.e. actual price difference is equal to their matched price
difference), we assigned each floor an equal rank.
As expected, floor played a more prominent role in choice than in pricing. We first
compare choosers and price-setters best choice (rank one). Choosers assigned the rank of one to
a good floor 97.7% of the time (only one participant assigned a rank of one to a bad floor). In
contrast, price-setters (developers) were inferred to have assigned a rank of one to a good floor
only 36.2% of the time. (two-way chi-squared =37.62, p<0.001). These results replicated the
inconsistency between choice and pricing, as in study 4.
Next, we examine the entire range of ranks. We computed the average rank per floor
separately using the observed ranks, under choice, and the inferred ranks, under pricing. In
Figure 8, we report a box-plot with the mean rank and 95% confidence interval for each of the 10
- 24 -
floors separately for choice (upper panel) and pricing (lower panel). Interestingly, if we assume
that differences in floor ranks correspond to differences in the expected sales rates, we observe
roughly a good-faster-than-bad pattern for participants in the choice condition. In contrast, we
observe roughly a bad-faster-than-good pattern for developers in the pricing condition. Recall
that the prices under choice were simply the mean prices reported by the developers. Hence,
roughly speaking, the pricing strategies used by the developers appear to have generated the
same sales pattern amongst potential consumers as we observed in the field sales data in study 2.
However, the sales pattern we observed among the developers when using the means of their
stated prices mimics the desired sales patterns described by price-setters in study 1.
Study 5 confirmed the potential role of the prominence principle in housing demand in
Shanghai. Using actual potential buyers from Shanghai as well as actual price-setters, our
findings show the inconsistency between choice (buyers) and matching (sellers pricing
strategies). Furthermore, the use of 10-story building allowed us to replicate the inconsistency
between the desired bad-faster-than-good pattern of sellers and the actual good-faster-than-bad
pattern observed in the field.

Conclusion
Decades of judgment and decision-making research has generated a vast number of
intriguing anomalies. One of the most celebrated anomalies is the preference reversal between
choice and matching. Most anomalies in the judgment-and-decision-making literature are
demonstrated only using hypothetical scenarios or using real outcomes with very low stakes. The
present research relates the choice-matching preference reversal to a real-world problem
involving billions of dollars in one of world most active real estate markets.
- 25 -
Specifically, we have identified an inconsistency between Shanghai real estate developers
desired sales pattern and the actual sales pattern that emerged in the market place. Interviews
with developers showed that developers wanted to achieve a flat or a bad-faster-than-good sales
pattern. Furthermore, developers routinely used stated-preference data to learn consumer
willingness-to-pay for different floors and used this information to price in a manner that should
have achieved the desired sales pattern. Nevertheless, anecdotal accounts from developers and
actual sales data from selected Shanghai real estate developments clearly demonstrated a good-
faster-than-bad sales pattern.
To provide a potential explanation for the inconsistency, we conducted two controlled
laboratory experiments. The results supported our explanation, indicating that the difference
between choice mode and matching mode could generate the inconsistency between the desired
and actual sales patterns of apartments on different floors of a building. That is, the inconsistency
could arise from the incompatibility between the matching mode used by developers to measure
willingness-to-pay for floors and the choice mode used by consumers to buy apartments on
different floors.
To the extent that our explanation drove the observed inconsistency, developers were
systematically setting prices using a biased estimate of demand. According to the compatibility
principle, developers could potentially reduce this inconsistency by simulating the consumer
choice condition in the determination of prices. For instance, developers could ask potential
buyers to hypothetically choose between a unit on a desired floor and one on a less desired floor
in repeated iterations, and through these iterations, systematically vary the price difference
between the two units. For example, they could start with two extreme cases: Would you buy
Unit 201 or Unit 801 if 201 is $300,000 and 801 is also $300,000? Would you buy Unit 201 or
- 26 -
Unit 801 if Unit 201 is $100,000 and Unit 801 is $500,000? Assuming that respondents choose
Unit 801 in the first case and Unit 201 in the second, then the developers can gradually increase
the difference in price from the first scenario and reduce the difference in price from the second.
Through this procedure they could find a point where consumers are indifferent between the two
units.
Our findings for the context of real estate pricing are interesting in their own right given
the sheer magnitude of the Shanghai market. Nevertheless, we do not expect the good-faster-
than-bad sales pattern we identified in this research to be limited to the real estate market in
Shanghai. Almost all products involve a tradeoff between quality and price, and consumers often
consider quality as more important than price (Simonson & Tversky, 1992) and quality may
frequently take more weight than price in consumers choices (Hardie, J ohnson, & Fader, 1993).
To the extent that sellers rely on matching to set prices, we expect to observe the same pricing
and sales-speed phenomena in other product categories.

- 27 -
7100
7200
7300
7400
7500
7600
7700
7800
1 2 3 4 5 6 7
Floor
P
r
i
c
e

p
e
r

s
q
u
a
r
e

m
e
t
e
r

(
R
M
B
)

Figure 1: Average price of units on different floors in the four mid-rises in Study 2 Dataset 1.

- 28 -

0.0
5.0
10.0
15.0
20.0
25.0
1 2 3 4 5 6 7
Floor
M
o
n
t
h
s

Figure 2: Average market-clearing duration of units on different floors in the four mid-rises in
Study 2 Dataset 1.

- 29 -
4000
4200
4400
4600
4800
5000
5200
5400
5600
5800
6000
1 2 3 4 5 6 7 8 9 10 11
Floor
P
r
i
c
e

p
e
r

s
q
u
a
r
e

m
e
t
e
r

(
R
M
B
)

Figure 3: Average price of units on different floors in the seven mid-rises in Study 2 Dataset 2
- 30 -
0
6
12
18
24
30
36
42
1 2 3 4 5 6 7 8 9 10 11
Floor
D
a
y
s

Figure 4: Average marketing-clearing duration of units on different floors in the seven mid-rises
in Study 2 Dataset 2.
- 31 -
10000
10500
11000
11500
12000
12500
13000
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Floor
P
r
i
c
e

p
e
r

s
q
u
a
r
e

m
e
t
e
r

(
R
M
B
)

Figure 5: Average price of units on different floors in the high-rise in Study 2 Dataset 3.
- 32 -

0
3
6
9
12
15
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Floor
M
o
n
t
h
s

Figure 6: Average marketing-clearing duration of units on different floors in the high-rise in
Study 2 Dataset 3.
- 33 -

5200
5600
6000
6400
6800
7200
7600
1 2 3 4 5 6 7 8 9 10
Floor
P
r
i
c
e

p
e
r

s
q
u
a
r
e

m
e
t
e
r

(
R
M
B
)

Figure 7: Average price of units on different floors in the 10-story building in Study 5.
- 34 -




0
2
4
6
8
10
1 2 3 4 5 6 7 8 9 10
Floor
R
a
n
k
0
2
4
6
8
10
1 2 3 4 5 6 7 8 9 10
Floor
R
a
n
k

Figure 8: Average choice rank for units on different floors in the 10-story building in Study 5.
The upper panel is for the choice condition and the lower panel is for the pricing condition.

- 35 -
References
Auh, Seigyoung and Michael D. J ohnson (2005), Compatibility effects in evaluations of
satisfaction and loyalty, J ournal of Economic Psychology, 26(1), 35-57.
Carman, Ziv and Itamar Simonson (1998), Price-Quality Trade-Offs in Choice Versus
Matching: New Insights Into the Prominence Effect, Journal of Consumer Psychology,
7(4), 323-343.
Chernev, Alexander (2005), Feature Complementarity and Assortment in Choice, Journal of
Consumer Research, 31 (4), 748-759.
Economist (2001), Sticky as Treacle, Finance and Economics, J une 14.
Economist (2005), Sky-High Shanghai, Finance and Economics, March 23.
Fisher, Gregory W. and Scott A. Hawkins (1993), Strategy Compatibility, Scale Compatibility,
and the Prominence Effect, J ournal of Experimental Psychology, 19(3), 580-597.
Grant, Andrew (2006), The New Chinese Consumer, The McKinsey Quarterly, Special
Edition, 4-6.
Grether, David M., and Charles R. Plott (1979), Economic theory of choice and the preference
reversal phenomenon, American Economic Review, 69, 623-638.
Hardie, Bruce G.S., Eric J . J ohnson, and Peter S. Fader (1993), Modeling loss aversion and
reference dependence effects on brand choice, Marketing Science, 12, 378-394.
Lichtenstein, Sharman, and Paul Slovic (1971), Reversals of preference between bids and
choices in gambling decisions, Journal of Experimental Psychology, 89, 46-55.
Lichtenstein, Sharman, and Paul Slovic (1973), Response-induced reversals of preference in
gambling: An extended replication in Las Vegas, Journal of Experimental Psychology,
1O1, 16-20.
- 36 -
Liu, Wenjing and Dilip Soman (Forthcoming), Behavioral pricing, in the Handbook of
Consumer psychology, Frank R. Kardes, Curtis P. Haugtvedt, and Paul Herr eds.
Mahwah, NJ : Erlbaum.
Nowlis, Stephen M. and Itamar Simonson (1997), Attribute-task compatibility as a determinant
of consumer preference reversals, Journal of Marketing Research, 34(2), 205-218.
Payne, J ohn W., J ames R. Bettman, David A. Schkade (1999), Measuring constructed
preferences: Towards a building code, Journal of Risk and Uncertainty, 19 (1-3): 243-
270.
Prommerehne, Werner, W., Friedrich Schneider, and Peter Zweifel (1982), Economic theory of
choice and preference reversal phenomenon: A reexamination, American Economic
Review, 72(3), 569-574.
Reilly, Robert J . (1982), Preference reversalFurther evidence and some suggested
modifications in experimental-design, American Economic Review, 72 (3), 576-584.
Simonson, Itamar and Amos Tverskey (1992), Choice in context: Tradeoff contrast and
extremeness aversion, Journal of Marketing Research, 29, 281-295.
Slovic, Paul. (1975). Choice between equally valued alternatives. J ournal of Experimental
Psychology: Human Perception and Performance, 1, 280-287.
Slovic, Paul (1995), The construction of preference, American Psychologist, 50(5), 364-371.
Slovic, Paul, Dale Griffin, and Amos Tversky (1990), Compatibility effects in judgment and
choice, in Insights in decision making: A tribute to Hillel J. Einhorn, Robin M. Hogarth
Ed. Chicago: University of Chicago.
Tverskey, Amos, Shmuel Sattach, and Paul Slovic (1988), Contingent Weighting in J udgment
and Choice, Psychological Review, 95(3), 371-384.
- 37 -
Ye, Xiannian (2004), China Real Estate Market, http://www.china-
window.com/china_market/china_real_estate/china-real-estate-market--5.shtml.