Household food insecurity is experienced when there is uncertainty regarding, or a disruption in, food intake or eating patterns by at least one member of a household due to financial constraints( Reference Bickel, Nord and Price 1 , Reference Coleman-Jensen, Nord and Andrews 2 ). As a condition aligned with insufficient household resources necessary to obtain food in socially acceptable ways, it is more common in households with lower education, lower income, reliance on social assistance, of aboriginal status, where there are small children and where households are lone-parent led( Reference McIntyre and Rondeau 3 , Reference Tarasuk and Vogt 4 ). Intra-household food provisioning protects some household members, such as children, from food deprivation while differentially exposing other members, such as mothers, to compromises in food quality or intake( Reference Badun, Evers and Hooper 5 – Reference Travers 10 ). Thus, as a household measure, food insecurity reflects an overall state of food access adequacy. Household food insecurity is of public health concern because of its association with poor physical and mental health among household members( Reference McLeod and Veall 11 – Reference Vozoris and Tarasuk 13 ) as well as indicators of household stress such as family dysfunction( Reference Hamelin, Beaudry and Habicht 14 ).
Women are more likely to report living in a food-insecure household than men( Reference Alaimo 15 – Reference Rose 18 ). This is not surprising, given that women compared with men are more likely to be poor( Reference Rose 18 – 20 ), to live alone with their children( 16 , Reference Nord, Hooper and Hopwood 21 , Reference Primus 22 ), and to be dependent on welfare( Reference Tarasuk and Vogt 4 , Reference Burnham 23 , Reference Siefert, Heflin and Corcoran 24 ); all factors which are related to increased levels of household food insecurity in both Canada and the USA( Reference Tarasuk and Vogt 4 , Reference Nord, Hooper and Hopwood 21 , Reference Kirkpatrick and Tarasuk 25 , Reference Ledrou and Gervais 26 ). According to Health Canada, in 2007–2008 for example, 7·7 % of Canadian households were food insecure, and the prevalence of food insecurity among households led by female lone parents was 25·0 %; twice that of households led by male lone parents (11·2 %) and four times that of households led by couples (6·3 %)( 27 ).
Household food insecurity is currently measured in Canada and the USA using the US Department of Agriculture's Household Food Security Survey Module (HFSSM)( Reference Bickel, Nord and Price 1 , Reference Frongillo 28 – Reference Hamilton, Cook and Thompson 30 ). The eighteen-question multi-stage module includes three items relating to the household as a whole, seven items related to the adults in the household and eight further items pertaining to children. Of note, the module does not distinguish which adults or which children in the household are food insecure. Although a few international studies have aimed to assess if females and males respond differently to questions in the HFSSM( Reference Coates, Webb and Houser 31 , Reference Hackett, Melgar-Quiñonez and Pérez-Escamilla 32 ), response differences by sex of the respondent were not interrogated in the extensive validation studies for the HFSSM in Canada and the USA( Reference Bickel, Nord and Price 1 , Reference Frongillo 28 – Reference Hamilton, Cook and Thompson 30 ). We wondered if increased rates of household food insecurity observed in women living in disadvantaged households were entirely explained by such circumstances, or whether socio-culturally prescribed gender roles such as ‘gatekeeper of the family meal’( Reference McIntosh and Zey 33 , Reference DeVault 34 ) might account for some of the differences, in which case a systematic reporting bias by respondent sex would be at play.
The nationally representative Canadian Community Health Survey (CCHS), which has included the HFSSM since 2002, provided an opportunity to interrogate this issue, because in households where both an adult male and female are present, selection of the respondent is randomized. This created a natural experiment in which the sex of the respondent was independent of the sociodemographic characteristics of the household for married/cohabiting households. Therefore, male–female differences in the population-level prevalence of food insecurity for such households would be attributable to respondent sex. The objective of the present study was therefore to investigate if there is a gender/sex-associated bias in the reporting of household food insecurity for married/cohabiting households in Canada and to contrast these results with household food insecurity reporting in households where the sex of the respondent is predetermined by household structure (e.g. lone mother-led).
Data and survey design
The CCHS is a survey conducted by Statistics Canada which collected data on a two-year cycle until 2008, after which it became a rolling survey with a similar sampling strategy. The sample represents approximately 98 % of the Canadian population aged 12 years and older( 35 ) and excludes individuals living on aboriginal reserves, those residing in institutions, and those in full-time service with the Canadian Forces. CCHS sampling follows a multistage cross-sectional design in which the dwelling is the final sampling unit( 35 ). Once a household is selected, the household member to be surveyed is randomly determined by Statistics Canada prior to the interview. Approximately half of all interviews are conducted by telephone and half are conducted in person. This complex design ensures that the resulting sample is representative of the underlying population and, in households with both an adult male and adult female, the sex of the respondent is randomly determined.
To ensure a large sample for the present study, we pooled CCHS 3·1 (60 910 males and 72 037 females) and CCHS 4·1 (60 027 males and 71 932 females). Cycles 3·1 and 4·1 of the CCHS were conducted in 2005/2006 and 2007/2008, respectively. In cycle 3·1 the inclusion of the HFSSM was optional and implemented in six of ten Canadian provinces. In cycle 4·1 the module was mandatory. In order to assemble households that were sociodemographically similar for male and female respondents, we limited our analysis to households dependent on employment or self-employment for their primary income source, whose reported annual household income was less than $CAN 100 000. We excluded respondents living with parents, those less than 18 years of age, and those whose marital status was unclear. To ensure that our results were not driven by regions whose food insecurity might be driven by geographic lack of food access and potentially challenge the generalizability of our results, we also excluded respondents living in the three northern territories (Yukon, Nunavut, Northwest Territories). Respondents with missing observations for food insecurity, income source and amount, or household composition were also excluded. This resulted in a final sample of 25 176 non-married (i.e. single/widowed/separated/divorced; 11 865 male and 13 311 female) and 40 014 married (19 261 male and 20 753 female) respondents.
The primary outcome of the study was household food insecurity as ascertained through the HFSSM. The HFSSM metric is based on a series of component questions, ten asked of all households and an additional eight in households with children, regarding the household's ability to meet its basic food needs. The questions referred to the 12-month period prior to the interview and specified lack of money or other resources as being the reason for difficulty in meeting food needs. Component questions were asked over three stages, with each stage increasing in severity. A respondent was asked questions in the second or third stage only if an affirmative answer was given to at least one question in the previous stage( Reference Bickel, Nord and Price 1 , 36 ). In the USA, a household is labelled as ‘food secure’ if the respondent answers affirmatively to no more than two of the component questions; as having ‘low food security’ if between three and seven affirmative answers are given in households with children or if three to five affirmative answers are given in households without children; and as having ‘very low food security’ if eight or more affirmative answers are given in households with children or if six or more affirmative answers are given in households without children( Reference Bickel, Nord and Price 1 , Reference Hamilton, Cook and Thompson 29 , 37 ). For our primary analysis, households were considered food insecure if they were identified as having low food security or very low food security; otherwise they were considered food secure.
Because a gender/sex difference in reporting might differ by the sensitivity of the classification system, we considered alternative metrics for household food insecurity for married households. The first was the Health Canada metric of household food insecurity which, while based on the same component questions as the HFSSM, explicitly analyses food insecurity pertaining to children separately from that pertaining to adults. The metric identifies a household as food insecure if an affirmative answer is provided for two or more of the ten adult/household questions or if an affirmative answer is provided for two or more of the eight child questions. Note that any household identified as food insecure under the HFSSM metric will be identified as food insecure under the Health Canada metric. The latter metric was also used to examine adult/household and child scales of food insecurity separately, potentially mitigating any measurement bias between households with and without children resulting from differences in component questions( Reference Coleman-Jensen, Nord and Andrews 2 , Reference Nord and Romig 38 ). As a further sensitivity check, we considered as separate outcomes three of the first-stage component questions, asked of all respondents before staging for children. This avoided a potential bias between households with and without children that could result from higher response burden on respondents in households with children, who are asked more questions. Specifically, respondents were asked whether statements regarding the household members worrying about running out of food, having insufficient food and having a compromised diet were ‘often’ or ‘sometimes’ (classified as affirmative) v. ‘never’ true.
Primary analyses focused on stratification by respondent sex and marital status (married or living common-law (hereafter referred to as ‘married’)) v. other living arrangements such as single, widowed, separated/divorced (herein referred to as ‘non-married’). Additional analysis stratified married respondents by highest level of respondent education (less than high school, high-school diploma, post-secondary less than baccalaureate, and baccalaureate or higher) and respondent age group (18–29, 30–39, 40–49 and 50–64 years).
Household characteristics used in the analysis included household income, presence of children, number of children under 15 years of age or under 16–17 years of age, number of adults in household, home ownership, highest level of education attained by any household member, province of residence, and urban or rural residence.
Data were analysed using the Stata statistical software package version 10. Weights reflecting the respondent's contribution to the population were used to correct for sample selection and oversampling, and bootstrap weights were used to estimate standard errors, both provided by Statistics Canada( 35 ). Household characteristics were compared by respondent sex and marital status using t tests and χ 2 tests as appropriate with a significance level of P < 0·05. Univariate logistic regression was used to calculate unadjusted odds ratios reflecting the association between respondent sex and household food insecurity and between household characteristics and household food insecurity. This was extended to a multivariate logistic regression to calculate adjusted odds ratios, simultaneously adjusted for respondent sex and household characteristics. For married respondents the logistic regression analysis was repeated, stratifying separately by household characteristics (the presence of children in the household, household income, home ownership, urban or rural status of the household) and respondent characteristics (respondent's age group, respondent's highest level of education).
For married respondents the analysis was repeated replacing the HFSSM classification outcome with the Health Canada metric for household food insecurity, the Health Canada adult and child scales of food insecurity and, in separate analyses, with each of the three first-stage component questions described above. Finally, we examined HFSSM severity measures using multinomial logistic regression to calculate a relative risk ratio (RRR) reflecting the change in the risk of the household being identified as having low food security or very low food security, relative to food secure, when the respondent was female v. male. Bivariate multinomial logistic regression was used to calculate the unadjusted relative risk ratio with 95 % confidence interval, and multivariate multinomial logistic regression was used to calculate the relative risk ratio adjusting for household characteristics. Ethical approval was received from the Conjoint Health Research Ethics Board at the University of Calgary.
Household characteristics by respondent sex and married v. non-married status
Estimated differences in mean and proportional household characteristics by respondent sex and married v. non-married status are reported in Table 1. For non-married respondents, the household was 3·97 (95 % CI 2·93, 5·01) percentage points more likely to be classified as food insecure when the respondent was female rather than male. Significant differences across household characteristics between females and males were observed for non-married respondents. These included lower mean and median household incomes, larger households with more children (both under 15 and 16–17 years of age) and adults, higher home ownership, urban residence, and higher levels of household education for female respondents. Non-married women were also less likely than non-married men to reside in Quebec and more likely to reside in Ontario. For married respondents, the household was 1·78 (95 % CI 1·11, 2·44) percentage points more likely to be classified as food insecure if the respondent was female rather than male. Other than a small difference in residence in the province of Manitoba, for all other household characteristics, married male and female respondents were statistically indistinguishable. This is consistent with the randomization of respondent sex and suggests that the sample of married male respondents and married female respondents is representative of the same underlying population of households.
Significant difference between males and females: *P < 0·05, **P < 0·01, ***P < 0·001; confidence intervals are based on bootstrapped standard errors.
† Means calculated using population weights.
‡ Bootstrapped χ 2 statistic was used to test the null hypothesis of equivalent distribution between males and females. For non-married households the corresponding P values for household education are P < 0·001 (non-married) and P = 0·316 (married), and for province of residence P < 0·001 (non-married) and P = 0·623 (married).
Household characteristics and food insecurity
Estimates of the unadjusted and adjusted logistic odds ratios for food insecurity by sex, married v. non-married status and household characteristics are reported in Table 2. Households with higher income, higher education, who owned their home and who lived in Quebec or Saskatchewan v. Ontario had lower adjusted odds of being classified as food insecure for both unmarried and married respondents. Households with children either under age 15 or 16–17 years were significantly more likely to be classified as food insecure, regardless of unmarried v. married status of the household; this may partially be attributed to the well-documented and significant measurement bias attributed to differences in component questions between households with and without children( Reference Coleman-Jensen, Nord and Andrews 2 , Reference Nord and Romig 38 ). The presence of an additional adult increased the odds of food insecurity when controlling for other household characteristics in married households. Food insecurity was not independently associated with urban v. rural residence.
Ref., reference category; BA, baccalaureate.
Significant difference: *P < 0·05, **P < 0·01, ***P < 0·001; P values based on bootstrapped standard errors.
Respondent sex and household food insecurity
The unadjusted odds of a household being classified as food insecure was higher when the respondent was female rather than male, for both unmarried and married respondents (Table 2). After adjusting for household characteristics, the association between food insecurity classification and respondent sex was near zero and statistically insignificant for non-married households (OR = 1·01; 95 % CI 0·86, 1·19) and remained significant in households of married respondents (OR = 1·46; 95 % CI 1·24, 1·72). For non-married respondents, higher household food insecurity for female respondents than male respondents can be almost entirely attributed to less advantageous economic and demographic characteristics and circumstances particularly associated with lone parenthood, as inferred by controlling for number of household members under 18 years of age. For married respondents, the observations based on the univariate statistics are confirmed; adjusting for differences in household characteristics does not explain observed differences in the reporting of food insecurity.
Stratification by household and respondent characteristics: married respondents
The increase in the odds of a household being identified as food insecure when the respondent was female v. male persisted when we repeated the analysis for married respondents stratified by various household and respondent characteristics (Table 3). In households both with and without children, both the unadjusted and adjusted odds of the household being classified as food insecure were significantly higher for female respondents and changed little with adjustment. For example, for households without children unadjusted odds were (OR = 1·41; 95 % CI 1·11, 1·79) and adjusted odds were (OR = 1·43; 95 % CI 1·11, 1·85). Similar results were found when we stratified by home ownership and urban or rural status of the household. When we stratified by household income, we found that the increase in odds of a household being classified as food insecure for female respondents was most pronounced for households reporting a total annual income of $CAN 30 000–49 999 and $CAN 70 000–99 999.
Logistic regression-adjusted OR adjusted for: household income, presence of children, number of children under 15 years of age or under 16–17 years of age, number of adults in household, home ownership, highest level of education attained by any household member, province of residence, and urban or rural residence.
Significant difference: *P < 0·05, **P < 0·01, ***P < 0·001; P values based on bootstrapped standard errors.
By respondent age group, sex-stratification differences in the adjusted odds ratio were most pronounced for ages 30–39 years (OR = 1·62; 95 % CI 1·26, 2·07). The differences also remained apparent for other age groups in the sample, although generally not statistically significant. By respondent education, the adjusted odds ratio was largest across respondents who reported their highest level of education as some post-secondary education (less than a baccalaureate; OR = 1·63; 95 % CI 1·30, 2·05) and a baccalaureate degree or higher (OR = 1·84; 95 % CI 1·14, 2·97).
Alternative metrics for household food insecurity: married respondents
For married respondents, the adjusted and unadjusted odds ratios for female respondents exhibited a similar pattern to results using the HFSSM when we repeated the analysis using the Health Canada metrics for household food insecurity (Table 4). The estimated adjusted odds ratio differed, but not significantly, between using the child scale v. the adult scale; however, both were large in magnitude and consistent with the HFSSM estimates (OR = 1·38; 95 % CI 1·19, 1·60 for adult scale; OR = 1·44; 95 % CI 1·09, 1·91 for child scale).
HFSSM, Household Food Security Survey Module; RRR, relative risk ratio.
Logistic regression-adjusted OR and multinomial logistic regression-adjusted RRR adjusted for: household income, presence of children, number of children under 15 years of age or under 16–17 years of age, number of adults in household, home ownership, highest level of education attained by any household member, province of residence, and urban or rural residence.
Significant difference: *P < 0·05, **P < 0·01, ***P < 0·001; P values based on bootstrapped standard errors.
†Relative to food secure.
A similar pattern was observed when the three first-stage component questions were examined for married respondents. For each question, the adjusted odds ratio was approximately equal to the unadjusted odds ratio. These estimates suggest that in married households, female respondents are more likely than male respondents to provide an affirmative response to each of the three first-stage component questions.
Severity of household food insecurity: married respondents
Table 4 also presents multinomial logistic regression estimates of relative risk ratios for the two levels of food insecurity relative to being food secure for female v. male respondents. The adjusted odds of a married household being identified as having low food security relative to being food secure was significantly higher for female respondents than male respondents (RRR = 1·34; 95 % CI 1·12, 1·62). Further, a household being identified as very low food security relative to being food secure was also more likely for female respondents (RRR = 1·93; 95 % CI 1·30, 2·86).
In our sample of Canadian households, stratified by married and non-married respondents, we found that a household is more likely to be classified as food insecure, based on responses to the HFSSM, if the survey respondent is female. For non-married respondents, after adjusting for household characteristics, the difference in household food insecurity classification according to respondent sex disappears. This suggests that the differential classification can be attributed to differences in household composition and socio-economic resources. In particular, we found that, on average, the households of female respondents had more children and lower household income than the households of male respondents, suggesting insufficient material resources to feed the family( Reference Ledrou and Gervais 26 , Reference Bauman 39 – Reference Tarasuk 41 ).
In contrast, in married households where the odds of household food insecurity were also higher when women were respondents, the difference between the sexes persisted after adjusting for both household and other respondent sociodemographic characteristics, including the presence of children. Different metrics used for household food insecurity measurement revealed similar differences and, on the whole, a similar magnitude of difference between female and male respondents. Thus we suggest that the difference between male and female respondents is real and not a result of confounding factors.
Two mechanisms may lead to differential survey responses by respondent sex. The first mechanism is that men and women may have the same objective information about the household's food security situation but report it differently. This may result if women exhibit greater sensitivity to household needs than men, supported by the observation that women exhibit greater concern than men for the well-being of others( Reference Croson and Gneezy 42 ). The second mechanism is that men and women have different information available to them when they report on the household food security situation. This could be the case if women either deprive themselves of food in a food-insecure situation to protect others, including husbands, from experiencing food insecurity, or if women are more fully aware of constraints on food in the household because of their role as principal food provider. Indeed, both scenarios have been well described in food-insecure households( Reference McIntyre, Glanville and Raine 6 , Reference Burnham 23 , Reference McIntosh and Zey 33 , Reference DeVault 34 ).
Beyond the gendered social role of managing household food security, there is evidence of gendered roles in married households that might also make women's reporting different from men's. Survey evidence for married households has shown that women are more likely to take care of household budgeting and purchases of food, clothing and other essential items( Reference Phipps and Burton 43 , Reference Woolley 44 ). As a consequence, they may have better information about household food needs.
Several lines of evidence also suggest that women in married households may experience inequitable access to household resources( Reference Phipps and Burton 43 , Reference Lundberg and Pollak 45 , Reference Phipps, MacDonald and MacPhail 46 ). Studies have shown that resource allocation within a family, especially for child care and food( Reference Phipps and Burton 43 ), depends on income distribution among husband and wife. Inequity within the household might explain why married women could report higher rates of food insecurity compared with married men.
Our analysis does not distinguish between these mechanisms, any or all of which may be important. The resulting gender/sex-based bias in reporting does have implications though for the measurement of population levels of household food insecurity. To illustrate, consider the following thought exercise. Our sample is representative of approximately 4·5 million married households in Canada. Based on the estimated rates of food insecurity reported in Table 1, 189 410 households will be identified as food insecure if all respondents are male and 272 600 will be identified as food insecure if all respondents are female, a difference of 83 190 food-insecure households. This difference may have important implications for perception of the problem and consideration of allocation of public resources to address it.
Because we used a working sample, we could not examine whether stay-at-home parents v. working parents report food insecurity differentially, after accounting for household characteristics, particularly income.
Our results rely on the fact that married or cohabitating male and female respondents statistically represent the same population of households. We could not examine the responses of females and males in the same household to definitively show differential reporting within households by sex of respondent. Further, we lacked the information to distinguish between the different mechanisms by which differential reporting takes place and could not draw conclusions regarding whether male or female respondents more accurately report household food insecurity.
Our results do not negate the possibility of a reporting bias according to respondent sex in non-married households. Because respondent sex was not randomized across these households, we could not rule out the existence of unobserved household characteristics, correlated with sex, that influence the reporting of food insecurity of the household. Rather, we conclude that, in non-married households, differences in household classification of food insecurity by respondent sex are at least primarily accounted for by economic and sociodemographic disadvantages among respondent women.
Conclusions and implications
Food insecurity provides a unique measure of a household's physical and socio-economic lack of well-being( Reference Tarasuk and Vogt 4 ) with implications for public health responses and advocacy. We have shown that for households of married respondents, differences in measured household food insecurity by respondent sex appear to be associated with the respondent herself, not the household per se. One important implication of this finding is that the HFSSM may produce different food insecurity estimates in married households depending on whether the respondent is male or female. Despite validity and reliability testing of the HFSSM across different settings, but not according to marital status and sex in Canada and the USA( Reference Bickel, Nord and Price 1 , Reference Hamelin, Beaudry and Habicht 14 , Reference Frongillo 28 – Reference Hamilton, Cook and Thompson 30 , Reference Frongillo, Rauschenbach and Olson 47 , 48 ) as we have, there may be a need to reconsider who is best positioned to answer the food insecurity questions in dual-parent households. The Food Security Supplement in the US Current Population Survey is introduced in cohabituating households by requesting the adult most knowledgeable about food shopping and meal preparation act as respondent( 49 ). We expect that this may mitigate reporting bias that arises from differential information between men and women, but a reporting bias may still exist if both men and women in the household contribute to food shopping and meal preparation; if men and women in a household take differential responsibility for these duties, i.e. she shops, he cooks; and if men and women in the same household report the same objective information differently.
Our results cannot distinguish between under-reporting by males v. over-reporting or accurate reporting by females, which offers an important avenue for future research. Even without knowing which sex more accurately reports the food insecurity status of the household though, it is desirable to have consistent estimates to ensure comparability across the population and over time. To ensure consistency in the measurement of population-level household food insecurity, standards should be adhered to regarding respondent selection. For example, in married households, always randomizing or always selecting a female respondent will ensure that population-level estimates remain consistent.
Source of funding: This study was partially funded by a Canadian Institutes of Health Research (CIHR) Operating Grant (MOP-89731). J.M. was funded by a Post-doctoral Fellowship from the CIHR Population Health Intervention Research Network. L.M is supported by a Research Chair in Gender and Health from the CIHR Institute of Gender and Health. Conflicts of interest: None. Authors’ contributions: J.M. conducted the analytical work on the study. L.M. oversaw the conceptualization and overall execution of the study. Both authors contributed equally to manuscript preparation. Acknowledgements: The authors thank J.C. Herbert Emery for valuable comments.