Prevalence of food insecurity and implications for health

As noted above, we estimate the impact of an income transfer on food insecurity within households. This is important because food insecurity is a widespread issue, especially among vulnerable subpopulations, with far-reaching consequences for health and well-being.

In the Canadian context, more than 12 percent of households experienced some level of food insecurity in 2017–2018, affecting approximately 4.4 million people (Tarasuk and Mitchell Reference Tarasuk and Mitchell2020). Moreover, there are notable disparities based on geographic, demographic, and socioeconomic characteristics. For example, compared to the national average in 2017–2018, the prevalence of food insecurity was higher among households in Nunavut (57 percent), Indigenous Peoples (28 percent), families with children (16 percent), single mothers (33 percent), single fathers (22 percent), households with less than secondary education (21 percent), those with low income (36 percent) and social assistance recipients (60 percent) (Tarasuk and Mitchell Reference Tarasuk and Mitchell2020). These estimates pertain to household food insecurity but, as described in the introduction, the experiences of adults and children within households are not necessarily equal. For example, using Canadian data from 2007 to 2018, Ovenell et al. (Reference Ovenell, Azevedo Da Silva and Elgar2022) find that shielding occurred in approximately 40 percent of food-insecure families with children.Footnote ¹

The health consequences of food insecurity extend to both adults and children in the household, even in the presence of shielding. Among adults, food insecurity is correlated with poor diet quality, low self-reported health and increased risk of chronic disease, including hypertension, heart disease, obesity, and diabetes (Vozoris and Tarasuk Reference Vozoris and Tarasuk2003; Kirkpatrick and Tarasuk Reference Kirkpatrick and Tarasuk2008; Tait et al. Reference Tait, L’Abbé, Smith and Rosella2018; Domingo et al. Reference Domingo, Spiegel, Guhn, Wittman, Ing, Sadik, Fediuk, Tikhonov, Schwartz, Chan and Batal2020). There is also evidence that food insecurity makes it difficult to manage chronic conditions and adhere to the necessary medications (Chan et al. Reference Chan, DeMelo, Gingras and Gucciardi2015; Men et al. Reference Men, Gundersen, Urquia and Tarasuk2019). In addition to physical health, food insecurity increases the risk of anxiety, depression, mood disorders, and suicide ideation in adults (Muldoon et al. Reference Muldoon, Duff, Fielden and Anema2013; Davison and Kaplan Reference Davison and Kaplan2015; Davison et al. Reference Davison, Marshall-Fabien and Tecson2015; Martin et al. Reference Martin, Maddocks, Chen, Gilman and Colman2016; Jessiman-Perreault and McIntyre Reference Jessiman-Perreault and McIntyre2017). There is also evidence that food-insecure adults are more likely to die prematurely than their food-secure counterparts (Gundersen et al. Reference Gundersen, Tarasuk, Cheng, de Oliveira and Kurdyak2018; Men et al. Reference Men, Gundersen, Urquia and Tarasuk2020).

Among children, food insecurity is associated with poor overall health and higher risk of chronic disease, including asthma, overweight, and hypertension (Kirkpatrick et al. Reference Kirkpatrick, McIntyre and Potestio2010; Dubois et al. Reference Dubois, Francis, Burnier, Tatone-Tokuda, Girard, Gordon-Strachan, Fox and Wilks2011; Ke and Ford-Jones Reference Ke and Ford-Jones2015; South et al. Reference South, Palakshappa and Brown2019). It also impairs behavioral and cognitive development (Melchior et al. Reference Melchior, Chastang, Falissard, Galéra, Tremblay, Côté and Boivin2012; Ke and Ford-Jones Reference Ke and Ford-Jones2015). Likewise, food insecurity is detrimental to mental health, increasing the risk of depression and suicide ideation in children (McIntyre et al. Reference McIntyre, Williams, Lavorato and Patten2013). Even when children are shielded from food insecurity, they remain at greater risk of poor mental health compared to those in food-secure households (Ovenell et al. Reference Ovenell, Azevedo Da Silva and Elgar2022).

Universal Child Care Benefit

Past studies have shown that income-based policy interventions – such as child benefits, public pensions, and social assistance – reduce the risk of household food insecurity in Canada (Ionescu-Ittu et al. Reference Ionescu-Ittu, Glymour and Kaufman2015; Loopstra et al. Reference Loopstra, Dachner and Tarasuk2015; Li et al. Reference Li, Dachner and Tarasuk2016; McIntyre et al. Reference McIntyre, Dutton, Kwok and Emery2016; Brown and Tarasuk Reference Brown and Tarasuk2019; Tarasuk, Li et al. Reference Tarasuk, Li, Dachner and Mitchell2019). We build on this literature, especially Ionescu-Ittu et al. (Reference Ionescu-Ittu, Glymour and Kaufman2015), by estimating the impact of the UCCB on food insecurity within households. The UCCB was implemented by the Canadian federal government in July 2006 to help with the cost of raising children (Schirle Reference Schirle2015; Lebihan and Mao Takongmo Reference Lebihan and Mao Takongmo2019). It was a taxable benefit of $100 per month ($1,200 per year) for each child under the age of 6. Families could decide how to spend the transfer. According to Schirle (Reference Schirle2015), the UCCB was sizeable, representing up to 18 percent of the annual cost of raising a child. We further note that the UCCB represented 17 percent of average food expenditures in 2006, which were $7,046 per household (Statistics Canada 2008). The UCCB was a universal transfer for families with children under the age of 6 because it was not income-tested and 99 percent of eligible families received it (Human Resources and Skills Development Canada 2010). Moreover, other income-tested transfers, such as the Canada Child Tax Benefit and social assistance, were not clawed back as a result of the UCCB. There were major changes to the eligibility criteria and amount of the UCCB in 2015, and it was replaced by an income-tested child benefit in 2016. These changes are extraneous to our study period, which covers the period from 2005 to 2012.

Data source and sample

We use cross-sectional microdata from the public-use files of the CCHS, which is a nationally representative survey of the Canadian population aged 12 and older (Statistics Canada n.d.a).Footnote ² We pool observations from Cycle 3.1 (2005), 2007–2008, 2009–2010, and 2011–2012. We cannot use earlier data because measures of food insecurity are not comparable; earlier measures do not permit an examination of food insecurity within households or by degree of severity. Another limitation of the CCHS is that questions about food insecurity were “optional content” in Cycle 3.1 (2005) and 2009–2010 (and thus only asked to respondents in provinces and territories that selected them). Questions about food insecurity were “core content” in 2007–2008 and 2011–2012 (and thus asked to respondents in all provinces and territories).Footnote ³ Our sample is limited to provinces that were included in all years, such that they opted into the food insecurity questions in both Cycle 3.1 (2005) and 2009–2010, in addition to being observed when these questions were “core content” in 2007–2008 and 2011–2012. These provinces include: Nova Scotia, Quebec, Ontario, Alberta and British Columbia. We later relax this restriction as a robustness check, such that we include all provinces where questions about food insecurity were asked at some point during our study period. This includes provinces that only selected the questions in one of the two cycles in which they were “optional content” and provinces that never opted in – all of which are also observed when the questions were “core content”.

Our sample consists of households with children younger than 12 years. Questions about food insecurity are asked to a household representative, and we focus on those who identify as a parent living with one or more children. Further, following past studies on the UCCB (Schirle Reference Schirle2015; Koebel and Schirle Reference Koebel and Schirle2016; Lebihan and Mao Takongmo Reference Lebihan and Mao Takongmo2019), we restrict our sample to parents aged 25 to 49 as they are most likely to have children under the age of 6. This allows us to more accurately identify parents who were impacted by the policy change, as opposed to those who have young children in the household but did not receive the transfer – such as older individuals who reside with an adult child and grandchild(ren). We later relax this restriction as a robustness check.

We consider households observed in Cycle 3.1 (2005) to be in the pre-policy period, and those observed in 2007–2008, 2009–2010, and 2011–2012 to be in the post-policy period. In terms of whether households plausibly received the UCCB, we define the “eligible group” as those in which the youngest child is under the age of 6. The “non-eligible group” is defined as households in which the youngest child is aged six to 11 years. This is consistent with past studies assessing the impact of the UCCB on health and well-being (Ionescu-Ittu et al. Reference Ionescu-Ittu, Glymour and Kaufman2015; Daley Reference Daley2017; Lebihan and Mao Takongmo Reference Lebihan and Mao Takongmo2019). Our estimating sample consists of 46,210 respondents, of which 29,953 belong to the “eligible group” and 16,257 belong to the “non-eligible group.” Sampling weights are used in all analyses.

Dependent variables

Our dependent variables are derived from the “Household Food Security Survey Module” of the CCHS (Health Canada 2007). The module consists of 18 questions – ten are relevant to adults and eight are relevant to children. Household food insecurity is based on the experiences of adults and children therein. Refer to Table A.1 of the Supplementary Material the list of questions.

Consistent with past studies (Loopstra et al. Reference Loopstra, Dachner and Tarasuk2015; Li et al. Reference Li, Dachner and Tarasuk2016), we use the “Household Food Security Survey Module” to construct dichotomous measures of overall food insecurity (i.e., any degree) and moderate/severe food insecurity. In addition to the overall measure, it is important to consider moderate/severe food insecurity because of the greater negative impact on health (Tarasuk et al. Reference Tarasuk, Cheng, De Oliveira, Dachner, Gundersen and Kurdyak2015) and because interventions may affect food insecurity differently by the degree of severity (Li et al. Reference Li, Dachner and Tarasuk2016; Brown and Tarasuk Reference Brown and Tarasuk2019).

At the adult level, “overall food insecurity” equals 1 if there are any affirmative responses to the 10 questions pertaining to adults. It equals 0 otherwise. Likewise, “moderate/severe food insecurity” equals 1 if there are two or more affirmative responses to the adult-related questions, and it equals 0 otherwise. An analogous approach is used at the child level. Specifically “overall food insecurity” equals 1 if there are any affirmative responses to eight questions pertaining to children. It equals 0 otherwise. Likewise, “moderate/severe food insecurity” equals 1 if there are two or more affirmative responses to the child-related questions, and it equals 0 otherwise. Finally, at the household level, “overall food insecurity” equals 1 if adults and/or children therein have a value of 1 for this measure. Likewise, “moderate/severe food insecurity” equals 1 if adults and/or children in the household have a value of 1 for this measure.

Table 1 shows the prevalence of food insecurity at the adult, child, and household levels for the eligible and non-eligible groups, before and after the policy change. In the non-eligible group, the prevalence of food insecurity is consistently higher after the policy change. This is not observed in the eligible group, among whom the prevalence of overall food insecurity decreased at the adult, child, and household levels. However, there was an increase in the prevalence of moderate/severe food insecurity among children and households in the eligible group after the policy change relative to the pre-policy period (i.e., from 3.56 to 3.79 among children, and from 8.16 to 8.29 at the household level). In what follows, we examine these differences more rigorously in the context of the income transfer.

Table 1. Prevalence of adult, child, and household food insecurity among eligible and non-eligible groups, before and after the policy change

Note: Sampling weights are used in all analyses. Standard errors are reported in parentheses.

Empirical strategy

As shown in Equation 1, we use a difference-in-differences methodology to estimate the impact of the income transfer on overall and moderate/severe food insecurity at the adult, child, and household levels, respectively.

(1) $$\eqalign { {\rm{E}}\left( {{Y_{ipt}}|Eligibl{e_i},Treate{d_{it}},{{\boldsymbol X}_{it}},{{\boldsymbol Z}_{pt}},{\boldsymbol Province}{_p},{\boldsymbol Year}{_t}} \right) & = {{\rm{\beta }}_0} + {\rm{}}{{\rm{\beta }}_1}Eligibl{e_i} + {{\rm{\beta }}_2}Treate{d_{{\rm{it}}}} \cr & + {\rm{\alpha }}{{\boldsymbol X}_{it}} + {\rm{\gamma }}{{\boldsymbol Z}_{pt}} + {\boldsymbol Province}{_p} + {\boldsymbol Year}{_t} + {\epsilon_{ipt}}}$$

Y _ipt denotes food insecurity for household i in province p and year t (i.e., dichotomous measures of overall and moderate/severe food insecurity, respectively). Eligible _i is a dummy variable that equals 1 if the household had a child under the age of 6 and thus was eligible for the income transfer. Treated _it is a dummy variable that equals 1 if household i is expected to have received the income transfer in time period t (i.e., it indicates whether the household was eligible for the transfer in the post-policy period). X _it is a vector of respondent- and household-level covariates similar to those used in past studies (Ionsecu-Ittu et al. Reference Ionescu-Ittu, Glymour and Kaufman2015; Loopstra et al. Reference Loopstra, Dachner and Tarasuk2015; Li et al. Reference Li, Dachner and Tarasuk2016; McIntyre et al. Reference McIntyre, Dutton, Kwok and Emery2016; Brown and Tarasuk Reference Brown and Tarasuk2019; Tarasuk, Li et al. Reference Tarasuk, Li, Dachner and Mitchell2019). Specifically, as outlined in Table A.2 of the Supplementary Material, we control for the respondent’s age, sex, immigrant status, visible minority status, family structure (single parent versus two parents), household size, highest level of education in the household, income quintile, main source of income (employment insurance, social assistance or other sources versus employment income) and home ownership. Z _pt is a vector of province-level controls, including the annual average unemployment rate (Statistics Canada n.d.b) and all-items Consumer Price Index (Statistics Canada n.d.c). Province _p and Year _t denote province and year fixed effects, respectively.

We estimate Equation 1 using ordinary least squares regressions.Footnote ⁴ β₂ is the coefficient of interest (i.e., the difference-in-differences estimate), indicating the impact of the income transfer on food insecurity. It is identified on the assumption that, in the absence of the UCCB, the pre-policy difference between the eligible and ineligible groups would have persisted. This relies on the assumption that Treated _it is exogenous, in the sense that families were not sorting into the UCCB along some dimension that is also correlated with food insecurity. We argue this is plausible because the UCCB was a universal transfer for families with children under the age of 6; it was not income-tested and 99 percent of eligible families received it (Human Resources and Skills Development Canada 2010). As such, sorting into treatment would mean that the transfer induced families to have children. It is unlikely that the UCCB induced changes in fertility because it was small relative to the cost of raising an additional child (Daley Reference Daley2017) and, as reported by Schirle (Reference Schirle2015), there was no break in the Canadian fertility trend before and after the policy change.

Robustness checks

We conduct several robustness checks to examine the sensitivity of our main estimates to changes in the sample and specification.Footnote ⁵ First, we consider the possibility that other child-related policies affected food insecurity during our study period (Schirle Reference Schirle2015; Koebel and Schirle Reference Koebel and Schirle2016; Daley Reference Daley2017). Most policies were previously established and unchanged during this time, including the Canada Child Tax Credit and National Child Benefit Supplement (implemented in 1998), weeks of paid parental leave through the Employment Insurance program (changed in 2001) and low-cost child care in Quebec (implemented in 2000 and changed in 2004). Moreover, to our knowledge, child-related benefits introduced during our study period were relatively small and paid to all families with children under the age of 18; they did not differentially target those with children under six (the UCCB-eligible group) or those with children aged six to 11 (the non-eligible group). Such policies included the Child Disability Benefit (implemented in 2006), as well as the Children’s Fitness Tax Credit and a non-refundable child tax credit (implemented in 2007). There were, however, two notable policy changes that warrant further investigation, namely the introduction of the Ontario Child Benefit in 2007 (i.e., a non-taxable, income-tested monthly benefit for low- and moderate-income families with children under the age of 18) and the Quebec Parental Insurance Plan in 2006 (i.e., a more generous replacement for federal maternity and parental benefits). To ensure these programs are not driving our results, we re-estimate the regressions after dropping families in Ontario and Quebec, respectively. We further test the sensitivity of our results to these and other time-varying provincial drivers of food insecurity by re-estimating Equation 1 for the full sample with the addition of province-specific time trends.

Second, we drop respondents who were surveyed during the Great Recession of 2008–2009 to assess whether the impact of the income transfer was different under adverse economic conditions. To do so, we drop respondents from 2007–2008 to 2009–2010 because the date of interview is not observed in the public-use files of the CCHS.

Third, recall that questions about food insecurity were “optional content” in Cycle 3.1 (2005) and 2009–2010, whereas they were “core content” in 2007–2008 and 2011–2012. Our main sample is limited to provinces that were included in all years. As a robustness check, we extend the sample to all provinces where questions about food insecurity were asked at some point during our study period. This includes provinces that only participated in the “Household Food Security Survey Module” in one of the two cycles in which it was “optional content” and provinces that never opted in – all of which are also observed when questions about food insecurity were “core content.”

Finally, following past studies on the UCCB, our sample is limited to parents aged 25 to 49 (Schirle Reference Schirle2015; Koebel and Schirle Reference Koebel and Schirle2016; Lebihan and Mao Takongmo Reference Lebihan and Mao Takongmo2019). However, Ionescu-Ittu et al. (Reference Ionescu-Ittu, Glymour and Kaufman2015) do not restrict their sample in this way. Since our study directly builds on Ionescu-Ittu et al. (Reference Ionescu-Ittu, Glymour and Kaufman2015), we remove the age restriction as a robustness check, to ensure it does not impact our results.

Heterogeneity analysis

To explore heterogeneity in the impact of the income transfer on food insecurity, we construct subsamples by family structure, education, and income. Specifically, we consider single-parent families versus two-parent families. We then stratify the sample based on the highest level of education in the household: secondary or less versus post-secondary. Finally, we compare households in the third quintile or lower versus those with higher income.