U.S. flag

An official website of the United States government

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Institute of Medicine (US) Committee on Health Care for Homeless People. Homelessness, Health, and Human Needs. Washington (DC): National Academies Press (US); 1988.

Cover of Homelessness, Health, and Human Needs

Homelessness, Health, and Human Needs.

Show details

BThe Methodology of Counting the Homeless

Charles D. Cowan, William R. Breakey, and Pamela J. Fischer

Although there has been great interest in the number of homeless Americans in the past decade, few rigorously designed censuses of homeless populations have been mounted. When counts have been attempted, they have been local in scope, and problems with the enumeration methods have not been widely discussed.

The impetus for determining the number of homeless people results largely from increased interest in the projection of service needs and the distribution of resources for the homeless. For example, the U.S. Department of Housing and Urban Development has conducted its own research on the need for emergency shelters; the National Institute for Mental Health administers a number of service programs and has funded several research studies on the demand for services by the homeless; and P.L.'s 98-151 and 98-181 charged the Emergency Food and Shelter National Board with the quick distribution of $40 million to supplement and extend emergency food and shelter services nationwide. In this last case, the distribution of the funds was determined by considering both the overall unemployment rate for an area and the total number of unemployed people within a civil jurisdiction. Although the Emergency Food and Shelter National Board recognizes that ''unemployment rates and numbers are not a totally valid surrogate for need or poverty" (1983), they could find no other data "which were current, uniform and available within the time frame."

Counting the homeless population is extremely difficult because of the lack of a clear definition of homelessness, the mobility of the population, and the cyclical nature of homelessness for many individuals. In addition, homeless people are often reluctant to be interviewed, and many of them remain invisible even to the most diligent of researchers. There is no uniform method for counting the homeless, and very few good studies have been done. Three approaches have been used: indirect estimation, single-contact censuses, and capture-recapture studies. Each method, while offering some benefits, suffers from certain technical inadequacies.

Indirect Estimation

The indirect method involves eliciting information from knowledgeable sources, or key informants, about the number of homeless people in an area or the number receiving services, including tallies of the number of people using shelters and other services and estimates of the number of people turned away or otherwise not receiving services. This type of study requires that each of the informants must define "homeless" according to standard criteria and report the number of homeless people encountered over the same period and that the service agencies must be exhaustively surveyed. Because different agencies or groups provide services for the same set of people, some allowance has to be made for double (or multiple) counting of individuals. Unduplication is extremely difficult and requires detailed knowledge of the area and the services under study.

The great advantage of indirect estimation is that it is the most economical method. Data collection in this type of study can be done by telephone or by letter with staff in service agencies and local government bureaus. In addition, publications and service reports that can be used as a base for the counts are often available from the agencies and bureaus. The major disadvantage of this method is its tendency to produce inflated estimates due to duplication. The necessity of reliance on the advice of key informants whose perceptions of the size and nature of the homeless population are biased by their own particular set of experiences and who may be unaware of the extent of the overlap in service utilization may also badly skew that population estimate.

Two much-quoted national studies have reported widely divergent estimates of the national homeless population as determined by indirect estimation techniques. Hombs and Snyder (1982) reported that "in 1980 . . . we concluded that approximately 1 percent of the population, or 2.2 million people, lacked shelter. We arrived at that conclusion on the basis of information received from more than 100 agencies and organizations in 25 cities and states. . . . It is as accurate an estimate as anyone in the country could offer, yet it lacks absolute statistical certainty." This number, despite the flaws inherent in trying to obtain a national estimate from such a small and disparate sample, was for some time the only number available nationally and so attained a certain level of currency.

The second national study, conducted by the U.S. Department of Housing and Urban Development (HUD) (Bobo, 1984), also used the indirect method. Five hundred knowledgeable observers were contacted to obtain local estimates of the number of homeless people in a sample of 60 urban areas. In addition, HUD researchers spoke with 184 shelter managers in a separate random sample of shelters, visited 10 metropolitan areas, and interviewed officials in all 50 states. The HUD report estimated that there were from 250,000 to 300,000 homeless people in the United States.

There were several major flaws in the design and conduct of the HUD research. The first problem was that the contacts made in each of the 60 metropolitan areas were done by "snowball sampling," in which the interviewers first contacted sample units (shelters) that were known to them, and then used the information provided by shelter operators and other knowledgeable people to get names of other shelters or locations not on the initial list. Repetition of this technique should eventually lead to a complete list of all shelters, but several interactions are needed to be certain that the list is complete. Furthermore, this method leads to a different probability of contact for each unit in the population, since the probability is a function of how frequently the shelter is mentioned.

Another problem is the lack of uniformity of response or coverage from the 60 metropolitan areas included in the survey. Many advocates and others deal only with homeless people in their own immediate area, and may have no direct experience with homeless people in other parts of the metropolitan area or a good measure of how much movement there is between areas within the city. Obtaining estimates from people who have studied the homeless population for a whole city may be no better, since their methodologies and definitions vary. The city of Baltimore provides an example, where estimates of the number of homeless people there have ranged from 2,000 to 15,000 (Baltimore City Council, 1983; Health and Welfare Council of Central Maryland, 1983, 1986). Other cities have similar ranges, depending on how the research was done. Aggregating these numbers for 60 metropolitan areas and then weighting the numbers up to the national level may only be expected to produce estimates with larger meaningless ranges.

Applebaum (1986) points out that another problem with the HUD study is that it used population data for Rand McNally metropolitan areas (RMAs), which include but are much larger than the cities named in the survey. HUD asked respondents about the numbers of homeless estimated for the cities, but applied the reported homeless counts to the whole RMAs. This would lead to an underestimate of the number of homeless in the entire RMA; summing these estimates and weighting them up by the ratio of the U.S. population to the population of the sample RMAs would lead to an underestimate of the size of the homeless population for the entire country.

Finally, many local studies, including some of those incorporated into the HUD (Bobo, 1984) and Hombs and Snyder (1982) estimates, fail to distinguish between "point" estimates, which deal with the number of homeless on a particular day, and "period" estimates, which attempt to give a measure of the numbers of homeless over a period such as a year. The problems in deriving accurate period estimates are much more complex than those in deriving point estimates; estimates of the two types should never be combined.

A good example of an indirect count conducted at the state level is that by the Health and Welfare Council of Central Maryland (1986). A list was developed of all shelter providers in the state, and data were collected from each one regarding the numbers of people sheltered on specific nights throughout the year. Where shelter providers did not keep precise records, they were asked to estimate as closely as possible. In this way an estimate of the number of sheltered homeless people was developed. Estimating the number of homeless people not sheltered presented greater problems. Here again, the investigators asked the informed service providers and other concerned agencies to estimate how many people in their jurisdictions were homeless but not in shelters on given nights. The responses provided very wide ranges of estimates, so the investigators devised weighting systems to be applied to the different counties depending on their levels of economic development and organization. They also employed expert informants to estimate the proportions of homeless people who may not ever come in contact with shelter providers and thus would not even be included in the unsheltered estimate. They developed adjustment factors based on these estimates, despite the fact that there was little unanimity as to the size of this hidden population. They expressed the view that their estimates probably were very conservative. In this way, they concluded that on an average night in 1985, 1,000 homeless people were sheltered in Maryland and that there were an additional 1,900 unserved homeless people. Of this total population of 2,900, 1,160 were in the city of Baltimore; the remainder were distributed throughout the state.

The report briefly mentioned another figure: 28,038 people "reported sheltered during Fiscal Year 1985." The report was careful to point out that this figure is not based on unduplicated data, but on reports from shelter providers about annual volumes of service. The wide difference between this number and the one-night estimate illustrates the hazards of accepting service provider data without very careful consideration of how the data were obtained or whether they represented point or period estimates.

Another statewide indirect count was that done by the Department of Social Services in New York State in 1984 (New York State Department of Social Services, 1984). One thousand agencies were contacted to ascertain whether they were shelter providers. Two hundred and fifty shelter providers were finally identified from this list. Data from these providers led to an estimate of 20,210 single persons and family members as the average nightly sheltered population. In order to allow for the numbers of homeless people outside the shelter system, they used ratios of shelter: street populations derived from studies in Boston and Pittsburgh to arrive at a total statewide estimate of 50,362. The authors acknowledge that the use of these ratios, derived from estimates obtained by different methods in very different settings, is questionable. Additional data supplied by the shelter providers as well as by hospitals, police departments, and other informants supplemented the counts, to give more information on the composition of the homeless population.

Single-contact Censuses

The single-contact census is a technique that has been used in cities to make estimates of the size of their homeless populations. The census is usually taken by teams of individuals in a clearly defined area where preliminary studies suggest that the largest proportion of the homeless population can be found. A screening questionnaire, or, at the very least, instructions for selecting individuals to approach, are given to the teams conducting the census. Under optimum conditions the census should be conducted in a single day, preferably during a time of day when the homeless people are most likely to be stationary, such as late at night. However, for practical reasons, many censuses of this type are conducted over a short period of time with some mechanism for recognizing and eliminating duplications.

The advantages of a single-contact census are twofold: It provides for direct contact, even if only by observation so that the possibility of counting individuals more than once is reduced, and there is greater assurance that the people contacted fit the study's definition of homelessness. In addition, demographics and other information can be obtained that may be crucial to determining the type and level of services that need to be provided for this population.

There are also two primary disadvantages of a single-contact census. The census provides a cross-sectional view of the population at a single point in time, but because the homeless population appears to be in a constant state of flux (Bachrach, 1984; Bassuk, 1984; Lamb, 1984; Fischer and Breakey, 1986), it is out of date almost immediately after it is taken. Moreover, it may poorly represent the true homeless population if taken at the wrong time. If, for example, the number of homeless people on the streets is reduced on the few days after welfare or various types of social assistance payments become available, the number of homeless people may be underestimated.

Another disadvantage of a single-contact census is that it is expensive relative to indirect estimation. It is necessary to use a team of enumerators to comb areas of the city where data are being collected. For reasons of safety, workers are usually deployed in teams of at least two people who are often accompanied by off-duty police officers. Staffing costs are thus quite high.

An excellent single-contact census of the urban homeless was conducted in Washington, D.C., by the Center for Applied Research and Urban Policy of the University of the District of Columbia (Robinson, 1985). The study was carefully designed, and its techniques and assumptions are carefully documented. The investigators counted all the residents of the various shelters in Washington on a specific night, July 31, 1985, and obtained counts of homeless people in hospitals and other institutions. They supplemented this with a search of other places on the streets where people may be found at night. The city was divided into 20 count areas, with an enumerating team assigned to each area. The enumerators worked in pairs; each pair included a research assistant and a person experienced in working with the homeless. The investigators recognized that a certain number of homeless people would fail to be counted either because their appearance was unremarkable or because they chose concealed locations in which to sleep. An intensive search was therefore made in one area of the city with an augmented team that included a police officer to judge to what extent the less intensive counts may have failed to find homeless individuals who were hidden from view. A series of five estimates were made, based on the direct counts and including corrections for the two sources of error, underenumeration because people were not identified as homeless and underenumeration because people were actively avoiding being counted. The estimates ranged from 4,347 to 7,152, with the highest value being 64 percent larger than the lowest value.

Other recent single-count censuses have been conducted in a number of cities by surveying homeless people at sites that provide services, such as shelters, soup kitchens, and social service departments (Brown et al., 1983; Chaiklin, 1983; McGerigle and Lauriat, 1983). However, with survey sites of this sort there is an increased risk of duplication. This risk can be minimized by including brief screening questions and by restricting the data collection activities to a relatively short period. Surveys at sites that provide services can also have the problem of being dependent on agency personnel who may abandon or ignore the data collection because it interferes with their provision of services.

Multiple-count studies expand on the single-count methodology by conducting counts at two or more points in time. These studies are designed so that the counts are combined to produce a single estimate. Such studies provide additional information about changes in the population over time, documenting seasonal and other variations.

A recent study of this type was conducted in Chicago in 1985 and 1986, by Rossi and colleagues (1986). First, all homeless people in shelters were counted. Then, in order to estimate the number of street people, a survey design was developed to sample blocks in the city where homeless people were expected to concentrate according to information obtained from police and other informants. These blocks were then surveyed by research workers accompanied by police officers. This process was repeated 6 months later. Despite much effort, the yield of homeless people on the streets that were counted by this technique was very low, with only 22 being identified on the first occasion and 28 on the second. Based on these institutional and street samples, estimates were derived for the total homeless population of Chicago. The estimates, 5,907 on the first occasion and 3,719 on the second, were widely criticized by people familiar with homelessness in Chicago as being much too low. Previous estimates, derived by indirect methods, ranged from 12,000 to 25,000 (Chicago Department of Human Services, 1983). Another finding that casts doubt on the conclusions of this study is that no children were included in the counts of people on the streets, though families with small children are believed to make up as much as 40 percent of Chicago's homeless population (U.S. Conference of Mayors, 1986). Applebaum (1986) points out that many of the homeless people contacted on the streets may have denied that they were homeless. It is amazing that in a sample of blocks identified as likely places to encounter the homeless, only 22 of 318 individuals encountered would be homeless in phase I of the study, and only 28 of 289 would be homeless in phase II. Rossi and associates (1986) admitted that even when the police officers who accompanied the interviewers were not immediately introduced, subjects were always able to identify them as police officers, and therefore, they started the interaction on a negative note. In addition, the teams conducted preliminary observations of the blocks before any formal screening started, thereby tipping off a naturally suspicious population to their presence.

Having two counts enabled the investigators to comment on the differences in the findings obtained in October 1985 compared with those obtained in March 1986. In view of the methodological problems described above, however, the validity of these conclusions must be held in question.

Another multiple-count census was done in Nashville (Wiegard, 1985), where the homeless were counted on four separate occasions (the first day of each season) over the course of a year. Because Nashville is a much smaller city, the elaborate sampling frame used in the Chicago study was not needed and the entire downtown area could be surveyed during a single night. Demographic distributions observed at different times were used to draw conclusions about the changing nature of the homeless population in Nashville. The study concluded that although the estimated numbers of homeless people varied relatively little, from 689 to 836, the ratios of homeless found in shelters compared with the homeless found on the streets varied with the seasons. During the winter the ratio was found to be 25:1, but in the fall the ratio was 5:1.

Such ratios have been a focus of interest in several studies, including the study done for New York State by its Department of Social Services described above. The HUD report used an estimate that the shelter to street ratio was about 1:2 (Bobo, 1984). This estimate was derived from ratios of 100:129 estimated for Boston (Boston Emergency Shelter Commission, 1983), 100:130 for Pittsburgh (Winograd, 1982), and 100:273 for Phoenix (Brown et al., 1983).

Freeman and Hall (1986) attempted to use a ratio of this sort based on a survey of about 500 homeless people in New York City to make generalizations about the national homeless population. Apart from the obvious criticism that there is no logical basis for generalizing from New York City to the country as a whole, the many problems with this study included the local and unusual nature of their survey sample and their failure to take into account the cyclical patterns of homelessness. In attempting to generalize from their ratios to the national level, they based their estimates on the flawed HUD data and failed to take into account the variability of street: shelter ratios described above for various cities. Their conclusions, therefore, must be interpreted with considerable skepticism.

Capture-recapture Methods

Capture-recapture methods go beyond multiple-count methods by matching data on individuals observed at two or more points in time. They thus permit certain conclusions about the movement of individuals in and out of the population, as well as statistics about the population from which the sample was drawn. Capture-recapture techniques involve matching observations of individuals made at each of two or more data collection periods. In wildlife populations, for which this technique was developed, captured animals were tagged for ready identification on recapture. Matching of homeless individuals is achieved by using a combination of name, Social Security number, birth date, sex, race, and other unique identifiers. In matching subjects from the first observation to the second, the resulting data can be tabulated as shown in Table B-1.

TABLE B-1. Observation of the Homeless in Two Periods of Time.


Observation of the Homeless in Two Periods of Time.

The values in Table B-1 represent counts of people observed at different times: N 1 represents the count of those obtained during the first data collection period, N 2 represents the count of those obtained during the second data collection period, and M represents the number matched, that is, the number observed both times. The only number missing from Table B-1 that cannot be easily calculated by subtraction is the number of people in the population not observed in either the first or second period.

Two estimates of the number of homeless people in an area are possible from Table B-1. The first assumes that the census was complete and that the missing cell (not observed in either period) actually should have an entry of one. This estimate of the total number of homeless would be calculated as N 1 + N 2-M (Equation B-1). This estimate would then be merely a lower bound to the actual number of homeless, since in reality no census is complete and there are hidden homeless who remain uncounted no matter how strenuous an effort is made.

A second estimate can be calculated from Table B-1 that does not assume that there are no hidden homeless, and this is the estimate by the capture-recapture method. This assumes that each data collection is imperfect, that there is some probability at each data collection that individuals will be missed, and that consequently there is some (unknown) probability that individuals will be missed both times. The estimate of the total number of homeless from a capture-recapture study can be calculated from the formula (N 1 x N 1)/M (Equation B-2). Capture-recapture estimates have been used for biometric applications for several hundred years, chiefly in making estimates of the size of wildlife populations, and the basic estimator (Equation B-2) has been rederived in several different contexts. One of the earliest use of capture-recapture techniques for human populations was for the evaluation of the completeness of birth and death records (Chandrasekar and Deming, 1949). The most common application currently is for the evaluation of population and agriculture censuses (Cowan et al., 1986). Also called dual-system estimation in this context, evaluations of censuses by capture-recapture studies have been conducted in the United States, Paraguay, Bangladesh, India, and other countries. The evaluation of the census and use of the capture-recapture method in Somalia is of particular interest, since 60 percent of that country's population is nomadic and, in this respect, is similar to a homeless population. Additional information on a population can be obtained from making more than two observations. In recent years, maximum likelihood techniques have been used to derive estimates for use in studies involving several sampling periods (Bishop et al., 1976).

There are two studies of homeless populations that make use of capture-recapture techniques. The first was a study of the number of homeless men in Sydney, Australia (Darcy and Jones, 1975). In that study of homeless men, three 1-day censuses were conducted at 25 locations including shelters, hospitals, clinics, and a jail, on June 30, 1971; October 13, 1971; and March 8, 1972. Using Equation B-2, three estimates of the number of homeless men were obtained by comparing the three sets of data, two at a time, with the following results: June to October, 3,025; June to March, 4,119; and October to March, 3,322. The authors used a related technique that makes use of information from all three data sets to yield an overall estimate of the number of homeless men in Sydney (3,200). They also estimated the average "birth" and "death" rates for men entering and leaving the homeless population to be 21 and 5 percent, respectively, indicating that the homeless population was increasing over the period of the study.

It should be noted from the estimates presented above that the longer the interval between counts, the higher the estimate. The authors noted that the intervals between censuses were sufficiently long to allow entry and exit from the homeless population, through moving in and out of Sydney, deaths, and so on, so that the numbers of matches were reduced. If shorter time intervals had been used, it might be supposed that the estimates would have been lower.

The other study that used the capture-recapture method was conducted in Baltimore in 1985 and 1986 (Cowan et al., 1986). Four pairs of dates were chosen in August and November 1985 and in February and May 1986 to reflect the four seasons; each pair was used for a capture-recapture estimate. Data were entered on separate computer files for the eight counts, and a computer program was written to match individuals between counts. Each of the four pairs of counts permitted a capture-recapture estimate of the total number of potential shelter users in the city on an average night in that month. The results indicated that the number of people in the shelter-using population did not vary significantly across the seasons, ranging from 874 to 1,022. The counts also showed that on all eight nights the shelter beds in the city were filled close to capacity. From the computer lists it was possible to create a master file of individuals observed at any of the eight sampling periods, including demographic information and a record of sample in which they were included. The master file included 2,102 people, of whom 66 percent were men and 34 percent were women. Analysis of the patterns of recurrence of individuals in successive samples provided information on the dynamics of people entering and leaving the homeless population.

In order to make an estimate of the proportion of homeless people who do not use the shelter system, a street survey was conducted in December 1986. People were questioned very briefly in places where homeless people congregate, but do not sleep, such as soup kitchens, day shelters, or on the streets. The brief questionnaire asked whether they had a place to live, and if not, did they use the shelter system, and if so, when was the most recent occasion. This information was then used to supplement the capture-recapture estimates. It was found that about three-quarters of those questioned were potential shelter users. Taking this into account, the capture-recapture estimates from this study provided estimates that were very compatible with those obtained by the indirect survey method in the Health and Welfare Council of Central Maryland (1986) study described above, in which the total number estimated for Baltimore was 1,160 and the sex ratio was 64 percent male to 36 percent female.

The most important difference between the capture-recapture technique and the two methods described earlier is that the capture-recapture technique is the only one that involves a statistical model to estimate the size of the unseen portion of the population. Single-or multiple-count censuses require some correction or expansion on the counts obtained, to allow for the hidden homeless who are not included in shelter counts, or in the case of the single-contact census, who may not even be picked up in a well-done street census. The correction factors used in most studies did not seem to be calculated under any rigorous statistical procedure but, rather, reflected the ratio the researchers wanted to obtain.

A statistical model involves a number of assumptions about factors that affect or, perhaps more important, that do not affect the data collection process. The most important assumptions in the capture-recapture method are listed below:


Clear definitions: Homeless people can be accurately identified.


Homogeneous observation probabilities: Each person has the same chance of being observed in a specific period.


Stability: The size and nature of the population does not change during the observation period.


Stationarity: The population does not move in or out of the study area during the observation period.


Independent captures: For the periods, the order interaction term (however defined) is zero; that is, even though a homeless person was observed at one period, it does not affect the probability that the person will be observed on subsequent occasions.


Data correctness: The information collected is accurate.


Complete response: Individuals or informants provide information that is complete enough to permit matching.


Matching correctness: Data records for the same individuals can be linked between observation periods.


Single observations: Individuals are observed only once at each data collection.


Known externalities: Factors that affect the data collection are known and can be accounted for, such as weather conditions and receipt of welfare checks.

Violations of these assumptions invalidate the model, causing the results to be biased (Cowan, 1982, 1984). More complex models allow for all exigencies, but more complex models require more data and may be impractical.


Although the existence of a sizable homeless population is beyond doubt and there is a consensus among knowledgeable people that the extent of the problem has been increasing in recent years, the everchanging and fluid nature of the homeless population presents great methodological challenges in obtaining an accurate measure of its size. A review of the methods for estimating the number of homeless people indicates that great caution should be exercised in interpreting any of the available data. Each of the methods that has been used has inherent biases. There is no national estimate that is based on a sound methodology and that is agreed to be accurate. Estimates prepared for individual communities or cities may be more accurate, but here also, careful scrutiny of methodology is required to assess such data adequately.

In order to advance research in this area, developmental work is needed in three specific areas:


Definitions must be developed concerning who is considered homeless. Agreement on a definition is vital if valid comparisons between studies are to be made. Subgroups, such as homeless families, should also be defined.


There is a need for more comparative research to determine better methodologies for studying difficult to find or difficult to enumerate populations. An example would be research in the use of network or multiplicity sampling for making estimates of the size of the homeless population in cities.


There is a need for more comprehensive capture-recapture models. Such models would permit data from several sources to be used and adjustments for missing data to be incorporated into the model.

Future research must pay very careful attention to the biases introduced by the different enumeration methods. Research teams must take advantage of the knowledge of people who are familiar with the homeless population in designing data collection techniques and in defining and identifying homeless people. Even with careful attention to methodological issues, it may not be practical or possible to develop a valid national estimate of the total number of homeless people. If, however, studies are carried out in cities and communities across the country using clear definitions and clearly defined methods, a composite picture may be built that will ultimately be more informative.


  • Applebaum, R. P. 1986. Counting the homeless. Paper prepared for the George Washington University Conference on Homelessness.
  • Bachrach, L. L. 1984. The homeless mentally ill and mental health services: An analytic review of the literature. In The Homeless Mentally Ill, H. R. Lamb, editor. , ed. Washington, D.C.: American Psychiatric Association.
  • Baltimore City Council. 1983. Report of the Baltimore City Task Force for the Homeless. Baltimore: Baltimore City Council.
  • Bassuk, E. L. 1984. The homeless problem. Scientific American 251(1):40-45. [PubMed: 6740307]
  • Bishop, Y. M. M., S. E. Feinberg, and P. W. Holland. 1976. Discrete Multivariate Analysis. Cambridge, Mass.: MIT Press.
  • Bobo, B. F. 1984. A Report to the Secretary on the Homeless and Emergency Shelters. Washington, D.C.: U.S. Department of Housing and Urban Development.
  • Boston Emergency Shelter Commission. 1983. The October Project: Seeing the Obvious Problem. Boston: Boston Emergency Shelter Commission.
  • Brown, C., S. McFarlane, R. Paredes, and L. Stark. 1983. The Homeless of Phoenix: Who Are They and What Should Be Done? Phoenix, Ariz.: Phoenix South Community Mental Health Center.
  • Chaiklin, H. 1983. The service needs of soup kitchen users. Baltimore: School of Social Work and Community Planning, University of Maryland. (Unpublished.)
  • Chandrasekar, C., and W. E. Deming. 1949. On a method of estimating birth and death rates and the extent of registration. Journal of the American Statistical Association 44:101-115.
  • Chicago Department of Human Services. 1983. Homelessness in Chicago. Chicago: Chicago Department of Human Services.
  • Cowan, C. D. 1982. Modifications to capture-recapture estimation in the presence of errors in the data. Paper presented at the meetings of the American Statistical Association, Biometrics Section, Cincinnati.
  • Cowan, C. D. 1984. The Effects of Misclassification on Estimates from the Capture-Recapture Studies. Unpublished Ph.D. dissertation. The George Washington University, Washington, D.C..
  • Cowan, C. D., W. R. Breakey, and P. J. Fischer. 1986. The methodology of counting the homeless. In Proceedings of the American Statistical Association, Survey Research Methods Section. Alexandria, Va.: American Statistical Association.
  • Cowan, C. D., R. J. Magnani, P. P. Biemer, and A. G. Turner. 1985. Evaluating Censuses of Population and Housing. Washington, D.C.: Bureau of the Census.
  • Darcy, L., and D. L. Jones. 1975. The size of the homeless men population of Sydney. Australian Journal of Social Issues 10:208-215.
  • Emergency Food and Shelter National Board. 1983. Program Description. A board established by Congress and the United Way. Alexandria, Va.: United Way of America.
  • Fischer, P. J., and W. R. Breakey. 1986. Homelessness and mental health: An overview. International Journal of Mental Health 14(4):6-41.
  • Freeman, R. B., and B. Hall. 1986. Permanent Homelessness in America: Working paper no. 2013. Cambridge, Mass.: National Bureau of Economic Analysis.
  • Health and Welfare Council of Central Maryland. 1983. A Report to the Greater Baltimore Shelter Network on Homelessness in Central Maryland. Baltimore: Health and Welfare Council of Central Maryland.
  • Health and Welfare Council of Central Maryland. 1986. Where Do You Go from Nowhere? Homelessness in Maryland. Baltimore: Maryland Department of Human Resources.
  • Hombs, M. E., and M. Snyder. 1982. Homelessness in America: A Forced March to Nowhere. Washington, D.C.: Community for Creative Nonviolence.
  • Lamb, H. R., editor. , ed. 1984. The Homeless Mentally Ill. Washington, D.C.: The American Psychiatric Association. [PubMed: 6479924]
  • McGerigle, P., and A. S. Lauriat. 1983. More Than Shelter: A CommunityResponse to Homelessness. Boston: United Community Planning Corporation.
  • New York State Department of Social Services. 1984. Homelessness in New York State: A Report to the Governor and the Legislature. New York: New York State Department of Social Services.
  • Robinson, F. G. 1985. Homeless People in the Nation's Capital. Washington, D.C.: Center for Applied Research and Urban Policy, University of the District of Columbia.
  • Rossi, P. H., G. A. Fisher, and G. Willis. 1986. The Condition of the Homeless in Chicago. A report prepared by the Social and Demographic Research Institute, University of Massachusetts at Amherst, and the National Opinion Research Center, University of Chicago.
  • U.S. Conference of Mayors. 1986. The Continued Growth of Hunger, Homelessness and Poverty in America's Cities: 1986. A 25-City Survey. Washington, D.C.: U.S. Conference of Mayors.
  • Wiegard, R. B. 1985. Counting the homeless. American Demographics 7(12):34-37.
  • Winograd, K. 1982. Street People and Other Homeless: A Pittsburgh Study. Pittsburgh: The Emergency Shelter Task Force.



Charles D. Cowan is chief statistician, Center for Education Statistics, U.S. Department of Education. William R. Breakey is director of the Community Psychiatry Program, Department of Psychiatry and Behavioral Sciences, The Johns Hopkins Medical Institutions. Pamela J. Fischer is assistant professor, Department of Psychiatry and Behavioral Sciences, The Johns Hopkins Medical Institutions.

Copyright © 1988 by the National Academy of Sciences.
Bookshelf ID: NBK218229


Related information

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...