Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
CITATIONS READS
26 27
4 authors, including:
William A. V. Clark
University of California, Los An…
255 PUBLICATIONS 5,536 CITATIONS
SEE PROFILE
1. INTRODUCTION
The literature on intraurban migration has been expanding rapidly (see [37]
for a recent bibliography). Although the literature is substantial, the actual
decision-making aspects of mobility and the spatial implications of those deci-
sions are less well specified than discussions of the reasons for the move.
*The research for this paper was parthlly supported by National Science Foundation Grant
SOC7727362. The authors particularly wish to thank the anonymous referee. Answering the
questions he raised gave us valuable new insights into OUT continuing research.
temporal and spatial aspects of the Brown and Moore structure. In this latter
specification, the migration process is broken into three parts (the development
of a desire to move, the selection of an alternative location, and the decision to
stay or move), which are viewed as interrelated. The evaluation of the present
residential location relative to alternative locations is incorporated into the
decision to stay or move. However, no specific model of the search procedure is
incorporated into the general model. The mathematical specification of the
verbal model essentially concentrates on listing the arguments of logit functions
related to the threefold decision process. The analysis of individual mobility is
extended to an aggregate mobility model, represented in terms of a linear
equation.
The preceding brief review of studies of intraurban migration indicates that
an adequate formal analysis of the decision-making process that both relates to
individual behavior and is operational is not available. In fact, a fairly thorough
literature search has revealed no papers relating actual household search
behavior to an expected utility/Bayesian theoretical framework, in which the
importance of uncertainty and attitudes to risk enter as signhcant components.
There is, however, a large body of literature dealing with optimal strategies
for sequential search. Much of this literature, however, is directed to the
problem of job search and the question of unemployment. Some early work in
this area was by Stigler [44], who assumed a decision process in which an
optimal number of job vacancies is selected for inspection. McCall’s later work
[20] emphasized the sequential nature of search and the existence of reserva-
tion wage levels, above which a job offer would be accepted. Further studies
(e.g., Kohn and Shavell [19]) emphasized the duration of search in relation to
personal characteristics, such as risk aversion, and other work has concentrated
on the information collecting aspects of search (see De Groot [9], Rothschild
[35]). Although these studies provide some general insights into search proce-
dures, the major aim of the researchers has generally been to obtain fairly
general statements concerning such features as aggregate unemployment. The
models are not generally expressed in a manner to permit calibration or the
testing of derived hypotheses. Applications of these models to housing market
situations are less common, although Nichols [27], Ionnides [18],and Flower-
dew [ 101 have specifically considered sequential search in housing markets.
However, none of these authors considers models that could be operationalized
easily for empirical research. In the models we develop below, we have
employed some of the concepts from the theory of optimal sequential search in
the face of uncertainty. These models are then used to generate testable
hypotheses concerning market search.
THE MODEL
2The so-called expected utility hypothesis provides the most generally accepted approach to the
problem of making decisions under uncertainty. There is growing evidence that it has a powerful
explanatory role in many choice situations, and even that people who behave in a manner consistent
with this theory are the most s u d in their chosen careers (see [=I).
3We may also view the difference between the utilities of situations with and without search as a
measure of relative attractiveness rather than stress, depending on one’s taste for the alternative
expressions.
8 / Geographical Analysis
area is relatively homogeneous in terms of variation between houses, this may
be an acceptable assumption. If, as is usually the case, the selection of a
vacancy to inspect precedes the journey to view the vacancy, our model of
search evidently corresponds to Schneider’s [36] case of route-finding search. It
is also quite probable that the list of vacancies in the area to be searched is
filtered, and that the choice of vacancies to be viewed is obtained from the
filtered set. Such a filtering process is often carried out with the aid of a realty
agent (see [17]). It does mean, however, that the selection of vacancies to
inspect is a two-stage process. In the following model development, the first
stage of selection is not considered. Given that we can correctly characterize
both the information updating procedures and the process by which a vacancy
is selected for inspection from the total list of vacancies in an area, the model
derived below allows one to predict, in principle at least, the probability of
rejection of a vacancy, and hence the probability of further search.
Although the verbal model indicates the important features of the process we
are attempting to model, it is necessary to develop a more formal presentation
of the ideas and to expand the concepts symbolically so that we can obtain
specific measurable concepts and testable hypotheses. Figure 1 attempts to
capture the most important aspects of the verbal model. We have included
symbolic notation that will be used in the formal development of the model. In
this way the schematic presentation acts as a transition from a general verbal
description to the formal symbolic structures.
the housing bundle are not common day-to-day events. These changes con-
stitute major decisions in the life of a household.
A more accurate description of the household’s decision process (at least in
the short run) is that for daily decision making the bundle of housing is taken as
fixed. The shopping trip to the supermarket does not involve a household in the
decision about the optimal housing bundle. These short-run decisions can best
be described by the maximization of a “conditional” utility function.
max 6 subject to
pi qi = Y - M .
The solution to the problem yields the set of conditional demand functions.
4n=4n(pl , . . . , p , , y - M l X l . . . ~ ) .
Substituting these quantities into the utility function yields
V=V(p,Y-M,X).
It is not difficult to generalize the following analysis to higher dimensional
vectors of characteristics, but it must be emphasized that we have begged the
important question of what set of characteristics constitutes a “home.”
The beliefs of the household concerning the housing situation, which may or
may not prompt the beginning of search, relate to three major sets of factors.
Smith, Clark, Huff, and Shupiro / 11
The first is the set of beliefs concerning the spatial structure of the housing
market, and as a first approximation, we model these beliefs by assuming that a
household views the area of possible dwelling sites as being partitioned into a
finite number of areas, or neighborhoods, in each of which housing characteris-
tics are relatively homogeneous.
The second set of beliefs concerns the nature of the housing possibilities in
each of the neighborhoods. We assume these beliefs to take the form of a
subjective probability distribution Fi(X , M ) that describes the joint distribution
of housing characteristics and cost in the ith region (in general, one would
expect a positive correlation between X and M ) . The final set of beliefs
essentially relates to the costs of searching for a new dwelling.
These costs include the direct costs of search and the opportunity cost of
search time. Probably more important, however, are the transactions costs that
are involved once a tendered offer has been accepted, and a final search "cost"
of some importance involves the probability that a potential dwelling (found in
previous search) may be bid upon by another searcher should a bid be delayed
in favor of further search.
In connection with the household's preferences, these beliefs determine
whether the household will search for a new dwelling. The decision is reached
by comparing the expected utility of the household's situation were search to be
undertaken, with the utility of the present situation with no active search. We
now formalize the expected utility of a situation with search in order to arrive
at the decision criterion. To this end, consider a household with the current
housing situation Ho = (Xo,Mo)giving rise to a utility V". Several outcomes are
possible if the household decides to search for one more period. First, the
household may find a dwelling ( X , M ) that gives rise to a higher (or a lower)
overall utility than the current situation. Second, it may find in either case that
the expected utility of searching for another period is greater than the utility of
the current situation (or the best alternative). Third, if the household has
searched in a previous period and found a good option but delayed a bid, then
in the course of searching for another period, it may lose the prospect to
another bidder. Fourth, the process of further search may generate information
concerning the market, and this of itself may prove valuable. Finally, the search
process may involve costs, and a bid, if accepted, will also incur transactions
costs.
Besides the present housing situation Ho (giving rise to utility V"), the
household may have a housing option H B = (XB,MB)giving rise to a utility
V B>V".Let q(V B )be the expected utility of search at time t in region i given
that the best option currently available is HB. If we assume that the household
attempts to maximize its expected utility, it follows that the criterion function,
determining whether a household searches or not at any period and in any
region, is determined by comparing the expected utility of search with the best
utility available for certain ( V B )
E:(V”)=p{ Jvomax[
--m Vo,E,,,(V”)]dG(u)+~mmax[6,E,+,(u)]dG(~))
VO
viewed and the best is selected. The form of the distribution function reflecting
the household's beliefs will account for such differences.
Given the form of the functional equation (2), and given that beliefs change
during search, (2) represents an extremely difficult expression from which to
determine an analytical form for E;( V"), and the costs of empirical verification
would probably exceed the benefits of such verification. Hence we now make
the first of two major assumptions. The first presumes that households have
quite limited horizons and ignore the expected utility of search in future
periods. In this case, (2) reduces to
[
E/ ( V" ) = p Vo/ vo dG(u )
-w
+ /V" 6dG(u ) ]
O0
and (1)remains as the stress criterion for search in region i . The assumption
asserts that people search as if the present search period were to be the last
period of search. At the end, however, they revise their information and
reconsider their "decision." Search occurs at each stage in the area of highest
(positive) stress. At any point the area of search may change as beliefs are
modified and as new best alternatives are found. Search continues until \k' < 0
for neighborhoods.
Several important deductions may be drawn from the model (l),(3).First it
may be shown (see [ 101 for example) that such models possess a critical level of
utility V" such that if a vacancy offering a level of utility Vc is inspected, then
search ceases. Search will also cease if V>V" (satisfactory) and continue if
V<V" (in the present case, it should be noted that since beliefs may change
with search, then V" may also change). Second, there is a relationship between
the stress measure \k and the probability of further search P. For simplicity
consider the case of a housing market consisting of one neighborhood or area.
One may prove that if there exists some factor a that increases (or decreases)
the stress, then the probability of search in the next period, given the oc-
currence of search in the current period, must increase (decrease). In symbols,
if P is the conditional probability of search in the next period given search in
this period, then
\k = \k( VB,a),
where V B is the utility of the best available alternative. The critical level of
utility V" is obtained by solving
\k( VB,a)=o
vc=\k-1(o,a)=VC(a).
Hence the manner in which the probability of further search depends on a may
be obtained from
Hence a P / a a has the same sign as aVc/aa;but from the defining equation for
vc,
or
Since it is immediate that stress is lower the better the best alternative
currently available (a\k/aV,<O),and since by assumption a\k/aa>O, it
follows that 3Vc/i3a> O and hence aP/aa >O. Hence, any factor increasing
stress increases the probability of further search. One may partially extend this
result to a composite market, if a increases the individual's stress level in all
areas of the market simultaneously. One may illustrate these ideas by consider-
ing specific factors that enter the stress criteria. For example, on obtaining the
derivative of (l),(3) with respect to the probability of losing an alternative p,
Smith, Clark, Huff, and Shupiro / 15
one may show that a ' k / a p <0. Hence it follows that as the probability of losing
an alternative increases, so the probability of further search decreases. More-
over, it was noted above that the direction of change in stress with respect to
some factor a is the same as the direction of change of the reservation level of
utility V" with respect to that same factor. Since one may show that the
reservation level of utility increases directly with both wealth and initial utility
level V", it follows that the probability of search P also increases with wealth
(ceteris paribus) and with the initial level of utility V". It is apparent from these
previous remarks, which comprise part of the comparative statics of the model,
that such relations between 'k and P represent testable propositions obtaining
from the model.
Iso-utility Lines
-Lines O f Equal
Quan t i t y Probobili t y
Density
of
Housing
X
L
Housing Cost M
FIG.2.
16 / Geographical Analysis
\ I I = p [ V ~ ( M o , X o ) J JMO(%F(m,x)+Jm
--m --m
rn JrnV(m+t,x)dF(m,x)
- 00 Mo(r) 1
v(Y - M - t,x)
- Vy”x(MB-M)(XB-X)+V:
(XB - XI2
2 ’
( Z o - X B ) , (ZB
-X B ) , J- JMo(
00
O0
X)
dF(m,x ) , J JM B ( x )dF(m,4,
O0
--OO
O0
( m- a o ) 2dF(m, x ) , J J
l-: JMI( x ) -w
O0 O0
ME(%)
( m- dF(m, x ) ,
J-: lM:( x)
( x - Xo)’dF(m,x ) , /- O0
O0
J
&(x)
O0 ( x - $)2dF(m, x ) ,
( x - X O ) ( m - Boo)dF(m,x ) ,
J-- /MI( x )
As before, if 9 >0 in any region, search will occur, whereas if \k < O for all
regions, the best option is‘cho&n. The first four terms in (T), , remesent attitudes
to risk and preferences measured relative to the best avahble home HB;
S =V,”/ V: measures the rate at which the household trades off house pay-
ments against housing quantity/quality, being essentially the magnitude of the
slope of the indifference curve in Figure 2. R, = - :( V,”,/Vy”)is the usual
Arrow-Pratt [27]measure of risk aversion. For risk-averse households Ry is
positive and increasing with increasing aversion to risk or with respect to
expendable income Y - M. The number A, = - :( Vz/ V,”)may be given an
analogous interpretation to the degree of risk aversion concerning Y, and may
be interpreted as the price of the variance in X for the household. The
parameter T = ( Vr”,/ V,”) is less easy to interpret and is ambiguously signed. It is
p r o w o n a l to the percentage change in the marginal utility of expendable
income for a 1percent change in the quantity of house about X,.
There are two important points concerning the four preceding determinants.
First, according to the von Neumann-Morgenstern theory of expected utility,
the determinants are scale free. This is because the indirect utility for money
function V( Y - M , X ) is unique up to a linear transformation, i.e.,
represents the same preferences as V. Since the determinants S,Ry, Rx, and T
are ratios of derivatives, they are independent of a and b and are hence scale
free, allowing comparisons to be made between households. Various techniques
are available for determining S, Ry, Rx, [151, although T raises some problems
18 / Geographical Analysis
at present. Current research in the Los Angeles housing market is directed to
obtaining measures of these parameters.
A second point is that, since \k depends on these parameters, it is possible to
characterize the probability of further search P in such terms. The method is
similar to that presented above. It is possible to represent the situation
graphically in the X-M plane, as in Figure 2. A given stress function leads to a
curve (relating to \ k = O ) describing the set of houses which, if found, would
leave individuals indifferent between searching further and not searching. If the
stress function is increased by a change in some parameter, the indifference
curve must shift upwards (from curve A to curve B), but the probability of
finding a home in the market with such characteristics must decrease. Hence, at
least if vacancies are inspected randomly, probability of further search in-
creases. Hence it is possible to obtain the comparative statics of the probability
of further search in terms of the individual characteristics S, Ry,Rx, and T. For
example, one may show that if two individuals have the same preferences over
housing bundles (i.e., the same indifference curves in the X-M plane) but
possess differing degrees of risk aversion (Rx,Ry) then a\k/aR<O and it
follows by the results of the previous section that aP/aR <0, or the probability
of further search decreases with risk aversion.
The remaining determinants all represent beliefs concerningthe nature of the
market. (MB + z - Mo) represents the difference in house payments between the
best option, if bid upon, aqd the present optiin. (X, - Xo) has a similar
interpretation. The term (Mo-M,) involves M, which is defined as the
expected cost of a k m e in the market given that it is preferred to the present
home Ho. Hence (Mo - M,) represents the difference between the conditional
expected cost and the cost cf the best alternative. The next three terms have a
similar interpretation, and X,, for example, is the expected quantity/quality of
house g i u a that it is preferred to H,. The next set of terms are merely beliefs
concerning the conditional variances and covarkgces in fie housing market.
For example, we may interpret J??,J~o~x~(x- Xo)(rn- Mo)dF(rn,x) as the
covariance between house quantity X and house cost M given that the house is
preferred to the present house Ho. As in the case of the risk-aversion measures,
techniques are available for eliciting such beliefs concerning means and vari-
ances [22].Again it is possible to obtain a comparative statics of the probability
of further search in terms of the parameters representing beliefs.
It is apparent that, with the exception of T, our approximate theory has
provided us with a “complete” set of determinang relating search behavior to
preferences, beliefs, constraints, and risk aversion, and that most of the determi-
nants are in principle measurable. In effect, the model represents a set of
testable hypotheses concerning the effects of the determinants of search.
An implication of the model is that it also represents a sequential decision
process, and hence must be tested in a sequential setting, rather than in an a
postiori manner with a single interview. Given the former mode of testing, two
problems become amenable to study. First, if one obtains empirical estimates of
the determinants in (7) at each stage, how useful is the criterion in determining
whether search will continue or stop at any point and where search will occur?
To test the model in this respect, it may be preferable to take a logistic
5complete, that is, with respect to OUT theory.
Smith, Clurk, Huff, and Shaptro / 19
CONCLUDING REMARKS
It is unnecessary to reemphasize the sigdicance of investigations of intraur-
ban migration. The decision to move and the decision of where to relocate are
still among the most important decisions made by households within the city.
Moreover, the implications of the decision range from the purely personal costs
and returns from a move to system-wide changes caused by the patterns of
population redistribution (and population redistribution is the result of the
aggregated decisions of many individual households).
The model of decision-making and search behavior of prospective migrants
that we have developed is designed to provide a clear explanation of the
relationship between the individual migrant’s preferences and perceptions, the
characteristics of the external environment in which the choice is to be made,
and the actual choice of a new residence. The development of a sequential
choice model that specifically incorporates measures of uncertainty is a sigdi-
cant step toward a model that is truly representative of the decision-making
process. The particular power of the model developed in this paper is also
derived from the fact that we have been able to derive and estimate specific
parameters to be tested with empirical data. In this sense the model goes
beyond the more conceptual models that were discussed and analyzed in the
literature review.
We now summarize the main implications of our derivation of the determi-
nants of search. First, it was observed that although an essentially normative
model of household behavior was used, the lumpiness of the housing market led
to a search process that could be described in terms of satisficing behavior. This
concept agrees with the ideas of Brown and Moore [ 5 ] .Second, the model gave
rise to a very natural concept of stress, or relative attractiveness, with positive
stress leading to search, which itself leads to the possibility of stress becoming
negative for any neighborhood in the area of search. Furthermore, a given
household may search but revert to its initial dwelling, or it may search and bid
upon a new residence, or it may not search at all. Hence the criterion makes it
perfectly clear that the decision to search and the search process cannot be
separated in any natural manner. Third, the model is fully “spatial” in the sense
that a stress criterion is defined for each neighborhood, with search occurring at
20 / Geographical Analysis
any time in that neighborhood giving rise to the highest positive stress. Fourth,
the “myopic” approximation to the model gave rise to a set of relatively simple
and measurable determinants of the search process that allow a testing of the
theory. In particular the measures relate to the preferences, constraints, beliefs,
and degrees of risk aversion characterizing the household, with the prefer-
ence/risk aversion measures being scale free.
Finally, by defining the conditional probability of further search, it is possible
to relate stress and further search. In particular, if stress increases, then the
indifference curve representing the set of “reservation” houses shifts in the
space of housing properties, but the actual distribution of houses does not
change. Hence, the probability of finding a house better than the reservation
house changes (and in fact the probability of further search increases if stress
increases). Hence any factor increasing stress increases the probability of
further search, and one may generate hypotheses relating search probabilities to
all the personal and market factors that enter the stress criteria.
Until the model proposed in this paper is tested with empirical data, it is
difficult to identdy with any possible policy implications. However, it is possible
to comment on the potential insights that the application of such a model to the
residential relocation process may elicit. First, the model suggests that we will
be able to say something about where households will search in the city. If this
is true it follows that we can say something about where they will relocate.
Given this information it may be possible to make statements about where the
different types of people with different housing needs will relocate and also
about general questions of population redistribution within the city. Second, the
repercussions of such population redistribution questions take us directly to the
issues involved in busing, the provision of public services, and the financing of
the metropolis. Our final aim is to have a model that will be a relevant
instrument for policy comment on population and housing within cities.
Finally, some fairly specific implications of the search process for dwellings
relate to the personal and societal costs attached to poor relocation decisions.
Many households who move within the city move again in a short period of
time after the initial move. This suggests, of course, the impact of a poor initial
relocation decision. We believe that these poor decisions are in large part due
to inadequate or inaccurate information concerning the distribution (both
spatial and temporal) of better alternatives. The model suggests ways in which
search procedures for prospective migrants can be identified and indicates the
kind of information that, if made available to a migrant, would lead to better
and more efficient relocation decisions.
LITERATURE CITED
Barresi, C. “The Role of the Real Estate Agent in Residential Location.” Sodologiwl FOCUS,1
(1968). 59-71.
G t t , F. Residential Search Behatior. Toronto: York University Research Monographs,
1973.
Borch, K. H. The Ecaomics of Uncertuinty. Princeton: Princeton University Press, 1968.
Brown, L. A., and D. B. Longbrake. “Migration Flows in Intra-Urban Space: Place Utility
Considerations.” A n d , Association American Geographers, 60 (1970), 368-84.
Smith, Chrk, HUH,and Shupim / 21