• Paula C. Barata, Department of Psychology, University of Guelph, Guelph, Ontario, and Women's Health Program, University Health Network, Toronto, Ontario; Donna E. Stewart, Department of Psychiatry, University of Toronto and Women's Health Program, University Health Network, Toronto, Ontario.

  • This project was partially funded by Canada Mortgage and Housing Corporation (CMHC) under the terms of the External Research Program. The views expressed are the personal views of the authors and do not represent the official views of CMHC. This project was also partially funded through a postdoctoral fellowship to the first author from the Women's Health Program at the University Health Network. We thank Laurie DeOliveira, the research assistant on this project, for collecting the data and Dr. Ian Newby-Clark for statistical advice.

Address correspondence and reprint requests to: Paula Barata, Department of Psychology, University of Guelph, Guelph, ON N1G 2W1 Canada. E-mail:


Individual battered women have reported experiencing housing discrimination, but the extent of this problem has not been examined. This research used two experiments and a survey to determine if landlord discrimination could keep women from accessing rental units. In Study 1, a confederate asked 181 landlords about the availability of a rental unit in one of three living conditions (shelter, friends, no mention of current living conditions) and across two scenarios (does or does not have a child). Rental units were almost 10 times more likely to be available in the control condition compared to the shelter condition, χ2(1, N = 181) = 8.624, p = .003, and these results were not affected by whether or not the caller had a child, χ2(1, N = 181) = 0.214, p = .644. In Study 2, the confederate was employed and left a message on 92 landlords’ answering machines in the same three living conditions. The hypothesized comparison between the shelter and the other two conditions combined was significant, χ2(1, N = 92) = 4.602, p = .032. Finally, in a telephone survey of 31 landlords, a substantial minority (23%) said they would not rent to a hypothetical battered woman. The results of our studies suggest that discrimination against battered women by landlords is a real problem that is likely contributing to the difficulties that women experience in finding safe and affordable long-term housing.

The impact of intimate partner violence on women's housing has been of central importance since the beginning of the battered women's movement (Morley, 2000). There are many reasons why a woman might lose her home suddenly due to abuse (e.g., fleeing for safety, evicted for noise, forced out by violent spouse). This loss results in an immediate need for emergency shelter, which has increasingly been met by an organized shelter system. In Canada alone, there are now over 500 shelters across the country (Code, 2003). Although the importance of shelters cannot be understated, women must eventually find long-term housing. If they do not, they are at risk for returning to the abuser (Du Mont & Miller, 2000; Somers, 1992; Weisz, Taggart, Mockler, & Streich, 1996) or becoming homeless (Baker, Cook, & Norris, 2003; Metraux & Culhane, 1999; Pavao, Alvarez, Baumrind, Induni, & Kimerling, 2007).

Research on long-term housing has largely focused on the limited availability of affordable rental units (Morley, 2000; Peters, 1990; Weisz et al., 1996). Although research in this area clearly recognizes that interpersonal violence occurs in all socioeconomic groups, a woman's financial situation is often compromised when she leaves an abusive relationship because financial abuse may have occurred in the relationship itself and because leaving is often expensive. Nevertheless, finding housing after abuse is particularly difficult for low-income women because their options are more limited. There is no question that in recent years Canada and other Western countries have experienced a crisis in affordable housing that is especially apparent for citizens in the lowest economic brackets. A commonly used benchmark for affordable housing is a cost-to-income ratio of no more than 30%. In a recent survey, nearly 58% of Canadian residents in the lowest income bracket spent above this benchmark 3 years in a row (Rea, Yuen, Engeland, & Figueroa, 2008), suggesting that there is a gross lack of affordable housing for low-income households. This deficiency creates a climate that is ripe for discrimination because the most affordable housing is scarce and limited to the most desirable tenants.

Housing Discrimination Against Battered Women

To our knowledge, the effect of landlord discrimination has not been systematically examined, but some battered women have reported experiencing housing discrimination. Battered women have cited being turned away from rentals as a housing problem (Baker et al., 2003), and in the United States battered women have successfully sued landlords for discriminatory eviction practices (Lapidus, 2003; Weiser & Boehm, 2002). In a particularly prominent case, a woman was evicted after being assaulted by her husband on the grounds that someone in her control (i.e., her husband) had inflicted injury on a tenant (i.e., the woman herself) (Lapidus, 2003). This case brought to light a number of other discriminatory practices by the company that managed the apartment complex, including a policy against renting to abused women. The gravity of this problem has resulted recently in the passing of the Violence Against Women and Department of Justice Reauthorization Act in the United States, which specifically prohibits the eviction of tenants based solely on domestic violence, although this restriction is limited to landlords receiving government subsidies (Whitehorn, 2007).

These cases have identified a potential problem, but the extent of the problem and the reasons why some landlords may actively discriminate against battered women are not clear. Discrimination is defined as a negative behavior that is directed toward members of a socially defined group (Schneider, 2004; Stephan & Stephan, 1996). Like all people, battered women belong to a number of different social groups and thus the question of why others might discriminate against them is complex and multifaceted. For example, many women leaving shelters are single mothers on social assistance (Weisz et al., 1996), so it is possible that assumptions about battered women may lead landlords to categorize them as “single mothers” or “unemployed.” This is a possibility we consider later, but first we argue that battered women may be stereotyped and discriminated against as a direct result of their abuse history.

Discrimination against “battered” women There is no one unifying stereotype of a battered woman, but there are a number of negative perceptions that persist despite significant public attention to the problem. For example, women are often seen as at least partially responsible for the abuse (Harrison & Esqueda, 1999) and for ending the abuse (Taylor & Sorenson, 2005). Additionally, attitudes that are clearly negative continue to be documented, even among professionals who should be familiar with the dynamics of intimate partner violence such as physicians (Garimella, Plichta, Houseman, & Garzon, 2000) and criminal justice personnel (Hartman & Belknap, 2003). Hartman and Belknap (2003) interviewed judges, prosecutors, and defense attorneys and found that battered women were often placed in a no-win situation in which women who wanted to drop charges were seen as “pathetic, stupid, or even deserving of the abuse” (p. 363), but that women who actively pursued charges were seen as “vindictive, crazy, or falsely charging domestic violence to meet their own selfish needs” (p. 363).

There are also concerns that mental health professionals have incorrectly portrayed battered women as passive, helpless, or even mentally ill (Gondolf & Fisher, 1988; Schneider, 1986). These critiques have largely come in response to increased use of the “battered woman syndrome” (Walker, 1984, 2000) to understand the dynamics of abuse and to defend battered women who kill their abusers. It is unclear whether or not the general public holds these views, although research examining mock jurors’ decisions in cases depicting abuse scenarios suggests that they do. When women are not portrayed as typical passive victims, they are held more accountable by mock jurors for the murder of their abusive partner (Russell & Melillo, 2006; Schuller & Rzepa, 2002) and male abusers are held less accountable for their abuse (Kern, Libkuman, & Temple, 2007). The conceptualization of battered women as helpless or passive is concerning because negative attitudes can still result from beliefs that elicit pity as opposed to contempt (Fiske, Cuddy, Glick, & Xu, 2002). For example, some landlords may feel sorry for battered women, but still consider them to be bad or risky tenants because of these negative perceptions.

For a battered woman to be discriminated against there must be an opportunity for discrimination and she must be identified as battered. Much of the work on discrimination and battered women has occurred within the context of the criminal justice system because, within this context, she is recognized as a battered woman. For instance, victims of intimate partner assault report more negative experiences with police, prosecutors, victim assistance staff, judges, and the criminal justice system overall than do victims of non-partner assault (Byrne, Kilpatrick, Howley, & Beatty, 1999). Applying for housing also provides an opportunity for discrimination. Battered women may try to conceal past abuse from a potential landlord, but it is likely to surface during the screening process. She may have to leave the shelter telephone number, have bad references (due to violence in the home, police presence, neighbors’ complaints, etc.), have poor credit ratings (due to the partner's control of the finances, etc.), have visible bruising, or have other problems that are largely due to the batterer's actions. Landlords can therefore infer that a woman is a victim of abuse and may discriminate against her on that basis. Additionally, some landlords conduct criminal record checks that produce the names of both perpetrators and victims (Lapidus, 2003) and may discriminate on that basis. However, we do not know the extent to which this kind of discrimination exists and keeps women from securing housing.

Discrimination against lone mothers and (other) poor women It is also possible that discrimination against battered women results more directly from assumptions about their economic situation or their status as lone mothers. There is a growing literature on discrimination against poor women and against mothers, especially mothers in the workplace (Correll, Benard, & Paik, 2007), but much less work has explored discrimination against women in the housing sector. Housing discrimination research has largely been gender neutral and has instead emphasized racial discrimination despite the fact that women are much more likely to file serious sex- and family status–discrimination charges against landlords and residential managers compared to men (Tester, 2007). Sexual harassment is an exception to the gender-neutral norm (Reed, Collinsworth, & Fitzgerald, 2005; Tester, 2008). This work has emphasized that poor women and women in low-income housing complexes experience overt discrimination (i.e., threats of eviction, actual eviction, and removal of essential services such as heat) when they resist a landlord's sexual advances (Tester, 2008). Housing discrimination against mothers is also a problem. Illegal practices such as advertising units as adult only, charging extra for children, and citing safety (e.g., the presence of a second-floor balcony) as a reason for disallowing children in rental units are still being documented (Tester, 2007).

The combination of poverty and motherhood can be particularly problematic because mothers on social assistance encounter blatant housing discrimination that prevents them from renting the most affordable accommodations and pushes them into undesirable and overpriced rental units (Centre for Equality Rights in Accommodation [CERA], 2003). An astonishing 74% of single mothers with children living in poverty who moved in 1991 found accommodations in the medium or most expensive segments of the rental market (CERA, 1998). It is, therefore, certainly possible that housing discrimination against battered women may be, at least partially, a result of assumptions about motherhood and/or unemployment.

Documenting Housing Discrimination

There are some methodological issues that merit consideration for research on housing discrimination. Researchers often use scale surveys of perceived discrimination, but reporter bias is a problem. Audit studies are quasi-experiments “in which trained pairs of observers or ‘auditors’ pretend to seek employment or housing in response to a newspaper advertisement” (Dion, 2001, p. 526). The auditors represent different conditions (e.g., people of different ethnic groups) within the experimental design, resulting in a more externally valid exploration of discrimination than survey studies. However, a problem with audit studies is that there is extensive contact between the auditor and the landlord because the two meet in person, which introduces a considerable amount of noise.

To address this problem, Page (1977) developed an experimental design that involved contacting landlords by telephone. In this design, the interaction between the landlord and the caller is usually no longer than several seconds and is thus thought to reduce exogenous influences (Page, 1983). Page has used this method to examine discrimination against mentally ill patients (1983), AIDS patients (1999), and lesbians and gay men (1998). To our knowledge, this method has not yet been used with answering machines, but leaving a message provides some advantages. In particular, more information can be conveyed inconspicuously before the landlord indicates whether or not the apartment is available. For example, it would be odd to provide a lot of information to a live caller landlord before allowing him or her to speak and disclose the availability of the apartment; however, it is not unusual to leave a long message on voice mail. In the current study this allowed us to control for employment. In addition, answering machines and voice mail are in much greater use today than when Page developed this method in the 1970s and it is conceivable that some landlords may specifically use them to screen calls.

Some degree of deception is normally involved in studies on discrimination because social desirability may inhibit participants from admitting to discriminatory behavior. Implicit in the use of deception is that the participants would view their actions as discriminatory and thus be less likely to admit to them. However, it is not clear whether most landlords consider denying a battered woman a rental unit as an act of discrimination.

Current Research

The current research used two experiments and a survey to examine housing discrimination against battered women. In Study 1, we used the telephone method developed by Page (1977) to call landlords and inquire about a rental unit. Three different conditions were enacted in which the confederate was (a) staying at a shelter, (b) staying with friends, or (c) did not mention her current accommodations. The friends comparison condition was chosen to ensure that simply stating one's living conditions could not in itself explain any differences found between the shelter and the no-mention control conditions. Additionally, staying with friends might be perceived negatively as an indication of housing instability and could thus serve as a more complete comparison. We hypothesized that landlords would be less likely to indicate that an apartment was available to women staying at a shelter compared to women staying with friends or women who do not mention their current living condition. The shelter, friends, and no-mention control conditions were repeated across two scenarios in which the caller did or did not have a child, completing a full 2 × 3 design. We hypothesized that having a child would result in fewer available apartments overall, and we hypothesized that having a child would magnify the effects expected in the first hypothesis (i.e., there would be an interaction between living conditions and whether or not the caller had a child).

In Study 2, we adapted the telephone method described above. The confederate enacted the same three conditions (without any mention of children), but she left a message on an answering machine or voice mail indicating that the landlord should call her back at work (thus controlling for employment). We hypothesized that landlords would be less likely to return a call and indicate that an apartment was available to women in the shelter condition compared to women in the friends condition or in the no-mention control condition.

As mentioned above, landlords may not consider refusing a rental unit to a battered woman as discrimination and thus may be willing to admit that they would not rent to a battered woman. Therefore, a survey of different landlords was included to assess their willingness to rent their advertised unit to a hypothetical battered woman.


Study 1: Live Caller (Simple and Child Scenarios)


A female confederate called landlords and asked whether or not the apartment they had advertised was still available (dependent variable = available or not available). Calls varied by the confederate's current living condition (independent variable [IV]= staying at a shelter, staying with friends, and no mention of her current living situation). These three conditions were repeated across two comparison scenarios (IV = simple and child) in which the confederate did or did not have a young daughter, completing a full 2 × 3 design.

In the simple scenario the caller dialogue was as follows for the shelter condition: “Yes I’m calling about the one-bedroom apartment you have advertised at [location/address] and, actually, Im staying at a shelter for battered women right now, but I’d like to move at the end of the month. Is the apartment still available?” In the friends condition the italicized dialogue above was replaced with “I’m staying with friends right now, but” and in the no-mention control condition the italicized dialogue was removed. In the child scenario the three conditions were repeated with the additional statement “with my little girl” after “I’d like to move” in each condition.


A total of 181 landlords were called: 91 in the simple scenario and 90 in the child scenario, with 30–31 landlords in each of the living conditions within each scenario. Overall, 50.9% of the landlords were female (n = 163 because gender was not always clear from the voice), and the mean monthly price of the rental unit advertised was $833.95 (Canadian dollars) (ranging from $525 to $1,200, SD = 128.06, n = 166 because the price was not always listed).


The selection of landlords followed the method described by Page (1977). Toronto Star newspaper advertisements for one-bedroom unfurnished apartments between $608 and $1,008 in Toronto, Canada, were used to select the landlords who were called. This price range was determined by calculating the average rent of a random sample of one-third of the units listed on the first day that calls were made and allowing for a range of $200 in each direction. The rent band of $608 to $1,008 was chosen to help avoid outlier rental units that would be very under- or overpriced for a one-bedroom apartment. We determined that a $400 range would be adequate by examining the rent prices for that day and noting that this did not overly exclude many units. On the first day, 480 one-bedroom unfurnished apartments were listed, and the mean for the sample of one-third (n = 160) of the units was $808. Note that only a small number of the units from the first day were selected and called because approximately 20 landlords were called each day. Occasionally a rental unit advertisement did not disclose its price but was still included if selected, and these units tended to be more expensive (when disclosed by the landlord during the debriefing).

Calls were made during a 2-hour time period early in the day to help ensure that the advertised apartment would be available. Landlords’ telephone numbers were systematically selected each morning by calling every 10th advertisement. To ensure that a landlord was never called more than once, a temporary file with landlord telephone numbers was created. Before each new call was made, a computerized search of the new number was conducted against all previously called telephone numbers. Note that landlords were selected for both studies and the survey in the same manner and calls were made for all three studies during the same time period.

The calls were randomized across conditions in both studies. When each call was made, one of two possible outcomes was immediately apparent. The call could be picked up by a live caller or by an answering machine/voice mail. When the landlord picked up the phone, one of the six conditions in Study 1 or the survey was chosen at random. Randomization was accomplished by having the data collection forms stacked in two randomized piles, one for live callers and the other for answering machines. When each call was made, the next sheet of paper on the appropriate pile was used. Each pile was made by photocopying a cover sheet for each possible condition, shuffling small piles from each condition on a large table (face down) and then conducting a final shuffle of all the cover sheets for each pile.

A female graduate student was hired for data collection and trained to make the calls. She read from a script depending on which cover sheet was next on the pile (as described in the Design section above). She recorded the landlord's responses as to whether or not the apartment was available (yes, no, not disclosed) on the cover sheet immediately upon hearing the response.

Landlords did not consent to participate. After the landlord indicated whether or not the apartment was still available, the confederate immediately disclosed that she was a researcher and explained the true nature of the call. Disclosure was conducted at the suggestion of our research ethics board and was done to ensure that landlords were not becoming overly disconcerted by our research method. We know that this short deception was not overly detrimental because when the true nature of the call was revealed, only two callers expressed concern. Both of these callers were satisfied when we offered to send them a written letter describing the research study and they did not withdraw their data from the study. Most callers either reacted with curiosity or with mild annoyance (hung up), much like the landlords who were called to participate in the survey study (described below).

Study 2: Answering Machine Message (Employment Scenario)


As with Study 1, Study 2 varied by current living condition (IV = staying at a shelter, staying with friends, and no mention of her current living situation), but rather than speaking directly to a landlord, the confederate left a message on the landlord's answering machine or voice mail indicating that the landlord should call her back at work. In the shelter condition, the message was: “Hi my name is Jane and I’m calling about the one-bedroom apartment you have advertised at [location/address]. Im staying at a shelter for battered women right now, but you can leave a message for me at my workplace at any time. The number is [phone number]. Please call and let me know if the apartment is still available. Thanks.” The italicized dialogue above was changed for the friends condition and deleted in the no-mention control condition as in Study 1.


In the workplace answering machine scenario, 92 landlords were called and 62 (67%) returned the call. The average rental unit was $819.26 (ranging from $600 to $1,150, SD = 120.43, n = 78).


The selection and randomization of landlords was done in the same way and at the same time as Study 1. When an answering machine was reached, landlords were randomly assigned to one of the three conditions and asked to call the confederate back at work.

As with Study 1, the same female graduate student read from a script depending on which cover sheet was next on the answering machine pile. The names of the callers were changed so that, when the landlord called back and left a message, we could identify the condition. Three common Anglo-Saxon names were used (i.e., Jane, Patricia, and Rebecca). The outgoing message on the answering machine directed callers to the appropriate voice mail box (e.g., “If you are calling for Jane, press 1,” etc.).

Landlords who left a message on the answering machine were called back and if a live caller was reached, the confederate disclosed that she was a researcher and explained the purpose of the study. If by the third call the landlord was still unreachable, the true nature of the call was not disclosed.



An additional set of landlords were called and asked to complete a six-question survey. The first three questions were designed to parallel the two studies: willingness to rent the unit to a battered woman, willingness to return a battered woman's call, and willingness to rent to a battered woman with a daughter. The last three questions asked about experiences with domestic violence and their perceptions of risk in renting to a battered woman. The responses to the following two questions are reported in this manuscript: “If a woman called you about a rental unit and indicated that she was currently staying at a shelter for battered women, would you be willing to rent the unit you have advertised to her?” and “Do you think that renting to a battered woman is a risk? If yes, what kinds of risks do you see?”


A total of 54 landlords were asked to do the survey and 31 agreed, which is a response rate of 57%. Of those that agreed, 55% were women and the average asking rent was $856.00 (ranging from $600 to $1,150, SD = 136.27, n = 29).


As indicated above, 32 landlords were called for the survey study and were interspersed with landlords in Study 1, but only 17 agreed to participate. Therefore, an additional 22 calls were made to obtain a final sample of 31 completed surveys. The caller asked landlords if they would be willing to participate in a telephone survey seeking “to better understand why battered women have difficultly securing rental units.” If they agreed, she asked the six survey questions, recorded their yes/no/not answered response, and wrote down their open-ended responses as close to verbatim as possible. The responses were short, taking approximately 2–5 minutes to answer all six questions. No additional prompts were used beyond the basic questions. Participants in the survey design consented to participate and were never misled about the purpose of the call.


Study 1: Live caller (Child and Simple Scenarios)

Expected cell frequencies were examined for all pairs of discrete variables. All cells had expected cell counts that exceeded five in the first cross tabulation when the availability of the unit was grouped only by living conditions; however, when the cells were further broken down to account for the simple and child scenarios, 50% of the cells had an expected cell count that was less than five, with the lowest being 3.6. Responses in which the units were unavailable were collapsed with undisclosed responses to avoid an additional reduction in expected cell frequencies (see Table 1). A sequential logistic regression analysis was conducted using availability of the rental unit as the outcome variable.

Table 1. 
Responses to Telephone Requests in Each Scenario
 Available n (%)Unavailable + undisclosed + call not returneda= total (n)Total (n)
  1. aThis is only relevant for the workplace answering machine scenario.

Simple Scenario
 1. Shelter24 (80%)5 + 1 = 630
 2. Friends26 (87%)3 + 1 = 430
 3. Control30 (97%)1 + 0 = 131
Child Scenario
 1. Shelter21 (70%)6 + 3 = 930
 2. Friends27 (90%)2 + 1 = 330
 3. Control29 (97%)0 + 1 = 130
Collapsed Across Simple and Child Scenarios
 1. Shelter45 (75%)11 + 4 = 1560
 2. Friends53 (88%)5 + 2 = 760
 3. Control59 (97%)1 + 1 = 261
Workplace Answering Machine Scenario
 1. Shelter11 (35%)3 +3 + 14 = 2031
 2. Friends18 (58%)0 + 5 + 8 = 1331
 3. Control18 (60%)2 + 2 + 8 = 1230

Stated living condition (shelter, friends, no-mention control) was entered alone in the first model to examine the main effect of this variable. This produced a significant improvement over the constant-only model, χ2(2, N = 181) = 13.336, p = .001, indicating a main effect for stated living condition. The second model examined the main effect of having a child by adding the scenarios variable (simple vs. child) to the model. This did not produce a significant improvement in the predictability of the model, χ2(1, N = 181) = 0.214, p = .644, indicating that there was no difference in apartment availability across these two scenarios. The final model added the interaction between the three stated living conditions and the two scenarios. Similarly, this did not produce a significant improvement in the predictability of the model, χ2(2, N = 181) = 0.753, p = .686.

Table 2 provides the regression coefficients, standard error terms, Wald statistics, and odds ratios for each of the predictor variables and their interactions. The main effect produced in each model is largely accounted for by a large and significant difference between the shelter condition and the no-mention control condition in the hypothesized direction (variables were dummy coded to contrast against the control condition). In the first model, the no-mention control condition was almost 10 times as likely to produce an available rental unit compared to the shelter condition. As expected, there was no significant difference between the friends condition and the no-mention control condition in any of the three models.

Table 2. 
Logistic Regressions Predicting Rental Unit Availability
ModelVariablesBSEWald StatisticdfpOdds Ratio
  1. Note. Control is coded as 0 in the conditions variable (throughout) for comparison.

Experiment 1: Sequential Logistic Regression
1Conditions  10.092.006 
Conditions (Shelter vs. control)2.286.7788.6241.0039.833
Conditions (Friends vs. control)1.360.8242.7251.0993.896
2Conditions  10.0922.006 
Conditions (Shelter vs. control)2.286.7798.6231.0039.840
Conditions (Friends vs. control)1.359.8242.7211.0993.893
3Conditions  7.3492.025 
Conditions (Shelter vs. control)2.5201.0925.3221.02112.429
Conditions (Friends vs. control)1.1701.1850.9751.3243.222
Conditions × Scenarios  0.7452.689 
Conditions (Shelter vs. control) × Scenarios−.5051.5600.1051.746.603
Conditions (Friends vs. control) × Scenarios.3591.6510.0471.8281.432
Experiment 2: Direct Logistic Regression
 Conditions  4.4612.107 
Conditions (Shelter vs. control)1.003.5293.5981.0582.727
Conditions (Friends vs. control)0.080.5210.0241.8781.083

In order to examine the univariate difference between the friends condition and the shelter condition, the first model was rerun with dummy coding to contrast against the shelter condition. Because the variables are the same, the results are the same as for model one above, but the shelter-versus-friend comparison can now be examined. The Wald criterion was not significant, χ2(1, N = 181) = 3.420, p = .064. Finally, the living conditions variable was contrast coded in order to compare the shelter condition to the other two conditions combined (i.e., coded as shelter =−2, friends = 1, no-mention control = 1) and a multinomial logistical regression was run on this variable. As expected, this model was significantly different from the constant-only model, χ2(1, N = 181) = 10.080, p = .001, and resulted in a significant Wald statistic, χ2(1, N = 181) = 9.687, p = .002. The odds ratio indicated that the apartment was 4.15 times less likely to be available in the shelter condition compared to the other two conditions combined.

Study 2: Answering Machine Message (Employment Scenario)

Responses in which the units were unavailable, undisclosed, and unreturned were collapsed in order to reduce the number of cells with expected cell frequencies of less than five. After being collapsed, expected cell frequencies were examined for all pairs of discrete variables, and all cells had an expected cell count that exceeded five. A direct logistic regression analysis was conducted using availability of the rental unit as the outcome variable and the stated living conditions variable as the predictor.

The model was not significantly different from the contrast-only model, χ2(2, N = 92) = 4.626, p = .099, and the hypothesized comparison between the shelter and the no-mention control condition was not significant, χ2(1, N = 92) = 3.598, p = .058. As expected, there was no difference between the friends condition and the no-mention control condition (see Table 2). As above, the logistic regression was rerun with dummy coding to contrast against the shelter condition. The difference between the shelter-versus-friend comparison was not significant, χ2(1, N = 92) = 3.118, p = .077. Contrast coding (which was the same as in Study 1) revealed a significant difference between the shelter condition compared to both the friends and no-mention control conditions. The model was significantly different from the contrast-only model, χ2(1, N = 92) = 4.602, p = .032, and the Wald statistic was also significant, χ2(1, N = 92) = 4.439, p = .035. The odds ratio indicated that the apartment was 2.62 times less likely to be available in the shelter condition compared to the other two conditions combined.


QSR Nudist was used to assist in the analysis so that common responses could be identified. The answers were categorized first by question and second by dominant response (i.e., yes/no). Individual answers were coded and are described in the approximate order in which they were most common. The results below are exhaustive and identify all the topics that the landlords mentioned in their responses.

Willingness to Rent to a Battered Woman

Twenty (65%) of the 31 landlords said they would rent to a battered woman. The most common explanation (nine landlords) for being willing to rent to a hypothetical battered woman was that choosing a tenant was based on the tenant's ability to pay. This point was often phrased as a condition upon which they would base their decision. For example, “Yes, as long as she has got money and can pay rent.” There was also mention of government programs that ensured her ability to pay (i.e., direct deposit) or provided support for women (e.g., good neighbor program). Other responses included simply needing tenants and/or the money, and not being biased (five landlords). Five landlords expressed sympathy, which sometimes reflected their own experience with violence and four were quite vague (e.g., not sure or don't see a problem).

Seven landlords (23%) readily admitted that they would not rent to a battered woman. Three landlords said that something dangerous or bad would be the result. For instance, “I have my own kids here and wouldn't want anything to happen.” Two landlords blamed battered women for their situation. For example, “She did something. Women choose their partners and she chose wrong and I don't want her.” Two landlords did not think a battered woman would be able to pay the rent, and two said that there was a specific policy against renting to battered women. One landlord expressed sympathy, but still indicated s/he would not want a battered woman as a tenant. For instance, “I give donations to shelters, but I’m not looking for this kind of tenant.” Four landlords would not give a definite answer and said they would need more information to make a decision or they expressed one of the fears listed above (i.e., inability to pay, dangerous situation).

Is Renting to a Battered Woman a Risk?

Of those 11 landlords who believed that renting to a battered woman was a risk, 7 saw the risk as being that the abusive partner could locate her and/or move in with her again. The actual risk itself was often unclear and worded in terms of “causing problems.” For instance, one landlord said, “The man could return to this apartment and could cause problems for her and other tenants. It's a hard question to answer. Unless she had a restraining order, there could be problems.” Six landlords also feared that the rent money could be in jeopardy. For instance one thought that women might “escape from the apartment” without paying the rent. One landlord indicated that all tenants, regardless of abuse history, were a risk.


Reviewing our findings, our primary hypothesis was supported. In Study 1, landlords were almost 10 times less likely to indicate that a rental unit was available to a caller staying at a shelter for battered women compared to the no mention control. There was also a significant difference between the shelter condition and the other two conditions combined such that the rental unit was less likely to be available in the shelter condition compared to the other two conditions combined. Importantly, there was no difference between a caller staying with friends and the control condition, which indicates that simply stating your living condition could not explain the finding. A different sample of landlords was surveyed about their willingness to rent to a hypothetical battered woman and this also provided some surprising evidence for overt discrimination. Over one-third of the landlords surveyed believed that renting to a battered woman was a risk, and 23% admitted that they would not rent to a battered woman.

Explaining Housing Discrimination Against Battered Women

The reasons for the discrimination are a little less clear; however, the results offer some direction in this regard. In Study 1, assumptions about motherhood were examined by including a scenario in which the caller clearly had a daughter. A sequential logistic regression showed that having a daughter had no overall impact on unit availability. In addition, there was no significant interaction effect. Thus there was no support for our hypothesis that having a child would reduce unit availability or our hypothesis that having a child would further reduce unit availability in the shelter condition compared to the other two conditions. One explanation for the null effect may be that landlords assume all female callers are mothers and thus the simple design was no different from the child design; however, this seems unlikely. A more likely explanation for the null finding is that our manipulation was not strong enough to produce an effect because it is no longer unusual in Western societies to be a single mother. Statistics Canada census data for the city of Toronto documents 115,200 lone-female-parent families compared to 472,410 married families, 61,555 common-law families, and 20,915 male-led families (Statistics Canada, 2006). In addition, lone female families have the lowest income ($35,176) compared to all other family types ($59,671) (Statistics Canada, 2006), and thus they are likely to be overrepresented in the rental market we targeted. Given the prevalence of single motherhood, having one young daughter in and of itself may not lead to housing discrimination, but it is important to note that the same may not be true of mothers with multiple children, boys, or older children. Nevertheless, the results suggest that assumptions about a battered woman's status as a single mother could not account for the difference between the shelter and the other conditions.

In Study 2, employment was expressed so that landlords could not make assumptions about unemployment. The hypothesis that landlords would be less likely to return a call and indicate that an apartment was available to employed women staying at a shelter compared to other women was only partially supported. The direct logistic regression was not significant, however, contrast coding revealed a significant difference between the shelter condition and the other two conditions combined. Rental units were 2.6 times less likely to be available in the shelter condition. This evidence provides some support for the assertion that housing discrimination against battered women extends beyond assumptions about unemployment.

Direct comparisons between the results of the two experiments should be made with caution because the data were collected differently and returning a phone call is not the same as providing a yes/no response to a live caller, but it appears that the results from Study 1 are stronger than those from Study 2. The logistical regression is significant in Study 1 and not in Study 2. Additionally, although the contrast-coding analysis was significant in both experiments, it was stronger in Study 1 compared to Study 2 (odds ratio = 4.15 vs. 2.6). Unfortunately, it is not clear whether the differences between the two experiments can be explained by differences in power, differences in method, or differences in assumptions about employment.

The inclusion of the friends condition also merits some consideration in a discussion on why housing discrimination against battered women was found. This condition was included to ensure that simply stating one's living conditions could not explain a difference between the no-mention control and shelter conditions, and as expected, there was no difference between the control and the friends conditions in either experiment. A secondary purpose for the inclusion of the friends conditions was to attempt a broad comparison for housing instability or general tenant undesirability. In choosing this condition, we attempted to find a comparison group that could also be described as living in a temporary accommodation, as this arrangement could indicate a risky tenant independent of abuse history. If the shelter and friends conditions were significantly different, this would have provided more evidence that abuse history, in and of itself, could explain the significant difference between the shelter and the no-mention control conditions. However, in both experiments the difference between the friends and shelter conditions was not significant. Therefore, it is still possible that the housing discrimination found could, at least in part, be explained by landlords’ perceptions of battered women as transient or risky tenants. The question then becomes, why would a landlord perceive a history of abuse as an indicator of risk? We return to this question shortly in our discussion of the survey results. Overall, the results suggest that there is discrimination against battered women as tenants. In addition, there was some evidence to suggest that this discrimination may extend beyond assumptions about motherhood and unemployment.

The open-ended results from the survey data are consistent with this conclusion and provide some other possible explanations for the discrimination. A few landlords clearly blamed the battered woman and saw her as undesirable regardless of whether or not the abusive partner continued to be in her life. However, most landlords who said that renting to a battered woman was a risk were concerned with undefined “problems” that might be annoying to other tenants or put others in the building at risk for harm. More research is warranted in this area. It is unlikely that these concerns stem from actual experiences because most landlords (84%) in the survey said they had never been made aware of family violence in their rental units. It is possible that some of these perceptions are being formed though sensationalized media cases that highlight the drama of the violence and ignore the mundane logistics that wear abused women down (i.e., securing long-term housing, finding employment, court cases that go on for years, etc.). Alternatively, perceptions about annoyances to other tenants could reflect a continued downplaying of the violence that is experienced within the family context. It is unlikely that stranger violence against another tenant would be classified as an annoyance. Concerns about her ability to pay the rent were also common, but these were not directly linked to welfare or unemployment, so it is not clear why landlords associated a history of abuse with an inability to pay the rent. One landlord indicated that a woman might have to run away and thus leave without paying the rent, but no one else mentioned this. Rather one is left to wonder whether these excuses serve as convenient covers for implicit biases.

The hidden issue might be that battered women are implicitly associated with incompetence. Fiske and her colleagues have found that dimensions of warmth and competence are particularly important in perceptions of stereotyped out-groups (Fiske et al., 2002; Fiske, Xu, Cuddy, & Glick, 1999). They argue that a stereotyped group is usually perceived as low in warmth and high in competence (i.e., an envious stereotype such as businesswomen) or high in warmth, but low in competence (i.e., a paternalistic stereotype such as the elderly). Some of the landlords in our survey may have held a paternalistic stereotype about battered women, which led them to generally view them as likable (some even expressed sympathy), but to still perceive them negatively as tenants. A perceived inability to pay the rent may provide a rationale for their negative perceptions and thus their expressed discrimination. However, more work is needed in this area to fully explore these possibilities.

It is worth noting that the candid answers we received from some landlords about their unwillingness to rent their rental unit to a battered woman surprised us. We would have thought that social desirability pressures would have muted this forthright disclosure. However, it is likely that these landlords did not view this as unjustified discrimination.

Impact of Housing Discrimination on Battered Women

Regardless of why housing discrimination exists, the resulting effect for battered women is very problematic and can be expected to impact a large number of women. Intimate partner violence against women has been increasingly well documented by nations and international bodies since the 1970s and a substantial number of women are affected. A review using population-based surveys from 36 different countries found lifetime prevalence rates for physical harm by an intimate male partner ranging from 10% to 69% (Heise, Ellsberg, & Gottemoeller, 1999). In Canada, where the current study was conducted, 7% of women reported experiencing physical violence at the hands of a male spouse in the previous 5 years (Canadian Centre for Justice Statistics, 2005). Many of these women will need to find new long-term housing. Unfortunately, this search can be difficult because of multiple and interacting problems that many abused women experience that drain their recourses and influence their ability to secure and maintain housing. For example, physical (Ellsberg, Jansen, Heise, Watts, & Garcia-Moreno, 2008; Plichta, 2004) and mental health (Afifi et al., 2009) are negatively affected; relationships with family and friends are strained (Riger, Raja, & Camacho 2002); employment is compromised (Browne, Salomon, & Bassuk 1999); and affordable housing is scarce (Rea et al., 2008). The current study suggests that housing discrimination can be added to this list.

Housing discrimination limits the number of units that are available when affordable units are already scarce. This is also a time when women need to find housing quickly and are already emotionally and financially drained. Additionally, the difficulties of finding appropriate housing are compounded by a need for both affordable and safe housing. The need for safe housing cannot be overemphasized because women can be at the greatest risk for serious injury after leaving an abusive partner (Campbell et al, 2003); however, the most secure and desirable housing may be the most prone to discrimination. Housing complexes in desirable neighborhoods or with security personnel could be the most difficult to get into because of discrimination. More research is needed to explore this possibility. A first step might be to look more closely at the type of housing that abused women are able to find and keep after leaving an abusive relationship compared to women separating for other reasons. However, the research should go further and directly explore the effect that housing discrimination might have for different types of housing in different neighborhoods and at different price ranges. Research findings that racial minorities and low-income families are pushed into undesirable and overpriced units are particularly concerning (CERA, 2003) and battered women may have a similar experience.

For many battered women, low-income housing is the only housing they can afford, but this type of housing may also put them at increased risk for victimization by landlords. Research on sexual harassment by landlords is still in its infancy (Reed et al., 2005; Tester, 2007, 2008). However, this type of harassment is particularly threatening and different from workplace harassment because it involves home invasion, the withholding of necessities such as heat and water, and the possibility of losing one's home (Reed et al., 2005). More research is needed on this very overt type of sexual discrimination because battered women may be economically vulnerable in the few months or years after they leave an abusive partner. There is also a need for research on different types of housing discrimination against battered women. Our study found that at the first step in securing housing (establishing availability of a rental unit) discrimination was a potential problem, but this may be only the tip of the iceberg because court cases alleging discriminatory housing practices against battered women have largely focused on eviction (Lapidus, 2003).

The results of this study add to the general literature on housing and domestic violence and to calls for increased government attention to second-stage housing for battered women (Menard, 2001; Morley, 2000; Scyner & McGregor, 1988). The ability to find low-income housing has been described as a “high-stakes game of musical chairs” (Rollins, Saris, & Johnston-Robledo, 2001, p. 277) to convey the fact that there are always more people in need of low-income housing than there are available units. In both Canada and the United States, special programs already exist that put battered women at the top of waiting lists for subsidized housing in an effort to deal with this problem, but more could be done to specifically target the problem of housing discrimination. Subsidized units could be made available in high-security buildings that offer amenities such as video surveillance and security patrol; procedures for reporting suspected discrimination could be put into place and explained to battered women and the service providers who help them; and landlords receiving subsidies from the government could be trained to better understand the plight of battered women and what constitutes discrimination. This last suggestion was one of the settlement requirements in a court case in the United States that found that a woman was unfairly evicted (Lapidus, 2003), but a strategy aimed at prevention would be more helpful than one aimed at litigation.

Little can probably be done to change the opinions of those landlords who hold openly hostile views toward battered women, and, fortunately, these landlords were in the minority. Those who hold vague assumptions about battered women bringing trouble or being unable to pay the rent might be encouraged to provide rental units to battered women if their concerns were addressed. Making positive connections between organizations that advocate for battered women and landlords might help change assumptions about women who have experienced domestic violence. Paradoxically, because landlords were quite open about their unwillingness or hesitation toward renting to battered women, they might also be willing to openly discuss the problem and help generate solutions.

Answering Machines in Housing Discrimination Work

In Study 2, we adapted a method that had previously been used to examine housing discrimination by including an answering machine scenario. This approach was necessary to control for assumptions about unemployment because employment could not easily and inconspicuously be conveyed in the live-caller scenarios. As such it is important to consider the validity of using an answering machine and the logistics of dealing with unreturned calls. A comparison of available rental units in the live-caller and answering-machine scenarios shows a clear difference. Across all three conditions there are more available units in the live-caller scenarios (87%) than in the answering-machine scenario (51%). In other words, researchers using an answering machine design can expect fewer available units and more unavailable units because of unreturned calls. This can be a strength in the design because expected cell frequencies may be more evenly distributed (as was the case here). However, it also means that researchers will need to consider how they will deal with unreturned calls.

Clearly, unreturned calls are not in and of themselves discrimination, but unreturned calls that vary by experimental condition might very well indicate discrimination. In fact, our study was inspired by unreturned calls. The first author was conducting a different study in a shelter for battered women that provided an opportunity for informal conversation as she waited for her next interview in the common area. Residents would often use the phone in that room to call potential landlords and would express frustration with answering machines because they would have to leave the shelter number and knew that their call would not be returned. Future research on housing discrimination should pay particular attention to answering machines and voice mail because landlords might use these devices to screen out some tenants through the use of call display or by paying attention to accents, tone, or caller availability (i.e., call after 5 PM or call me at work).

It is also becoming increasingly impractical to limit these kinds of studies to live callers because this systematically removes a large number of landlords. There are challenges with using both live-caller and answering-machine data, but there are also a number of benefits, including the potential for uncovering discrimination that is particular to answering-machine practices and the potential for results that are more generalizable.


There are some limitations in this study that should be considered in interpreting the results. In Study 1, cell frequencies in the logistic regression may have resulted in reduced power to detect differences between the simple and child scenarios. Whereas the expected cell frequencies were adequate in the first model, subsequent models exceeded the recommendation that no more than 20% of cells have an expected cell count of less than five (Tabachnick & Fidell, 2007). It should be emphasized that this does not detract from the significant differences that were found because Type I errors are not influenced by cell frequencies, but it reduces the conclusions that can be made about the impact that motherhood may have on housing discrimination and how this may interact with a previous history of abuse. It is worth noting that such low cell frequencies for unavailable apartments were unexpected because previous work by Page (1977, 1998, 1999) in various studies had found higher rates of apartment unavailability.

The most likely reason for this difference is that, when our study was conducted, rent control in Toronto had recently been eliminated. Previously, government policy had controlled increases in rent regardless of whether or not a tenant was new. This change in policy allowed for large increases in rent when an old tenant vacated. This scenario created a sharp increase in rent prices and a sharp increase in rental availability because many renters moved into a more affordable housing market. It is thus particularly relevant that discrimination was still occurring because high rates of available rental units should have discouraged discrimination.

Some bias may have been introduced by the confederate who was aware of the conditions; however, the interaction between the landlord and confederate before the landlord indicated whether the apartment was available was extremely short (generally less than 15 seconds), thus limiting the amount of bias that could be introduced. The answering-machine scenario holds even more promise for the reduction of bias because the caller can leave the same message on every machine without the possibility of unscripted dialogue between the confederate and the landlord.

The survey provided additional information about landlords’ perceptions. In particular, the open-ended responses provide some direction for future research, but it should be made clear that these were not in-depth qualitative interviews. The surveys were short and did not probe landlords to fully explain their answers or challenge them on their views. The responses were also not audio recorded because we believed that obtaining consent to record the interviews would have reduced landlord participation and candor because some may have been concerned with potential litigation if they admitted to discriminatory practices.

In Study 2, we chose to distinguish between landlords in the different conditions by providing each with a different caller name so that we could identify their experimental condition when they called back. However, this might have introduced a confound if the landlords had different impressions of the three names. We specifically chose names that were common, Anglo-Saxon, and appropriate for women in their 20s or 30s, but not too similar. Initially, we considered a very similar trio of names (Jane, Jan, and Jen), but decided that this might result in misidentified returned calls. We do not believe the chosen names affected the findings because the pattern of available units was similar in both experiments, but in the future we would suggest that researchers manipulate the telephone number rather than the name to avoid this problem.

It will be particularly important for future work to consider the possible interaction or additive effects of ethnicity and family composition on housing discrimination against battered women. In our study the confederate had no discernable accent so that she was presumably a Canadian-born woman, and we specifically chose three Anglo-Saxon names in Study 2 to avoid confounding minority ethnicity with experimental conditions. As such, our results may only generalize to some battered women and we cannot speak to the compounding effects of discrimination that many other women would face. It is quite likely, for example, that Black women face different challenges from White women. Tester (2007) reports that, although women file more sex- and familial status–discrimination cases against landlords compared to men, Black women report more than men and White women put together. Our research considered the effects of motherhood by including the child scenario, but the woman in this scenario was the mother of one young girl and this is likely to be different from being the mother of three or four children or of an adolescent boy.


Despite these limitations, our study contributes to an understanding of the housing challenges that battered women face when leaving an abusive partner and provides some direction for future research. It will be important to further explore the causes of the discrimination, the context in which it is most likely to occur (e.g., high-security buildings), the effect for battered women, and possible solutions to this problem.

Although housing discrimination against battered women had been reported in a few court cases, the extent of the problem was unknown. It was possible that court cases against discriminatory landlords reflected a few unusual cases. The results of our studies suggest that discrimination against battered women is real and more widespread than these isolated court cases would suggest. The findings here provide a starting point for a more nuanced understanding of how battered women are perceived and how this can lead to discrimination. Battered women face enough challenges; they should not be additionally burdened with discrimination that keeps them from renting the best housing that they can afford.