Children and Familial Economic Welfare: The Effect of Income on Child Development

5. Income and Children's Outcomes — Empirical Findings from the Reduced Form Regression Model

Our analysis in this section concentrates on analyzing two different reduced form regression models as well as regressions that investigate the effect of income on the home environment. Model 1 is a simple bivariate regression analysis of each of our four different income concepts by each child outcome. This model will serve as a baseline to study the effect that introducing controls has on the relationship between income and our selected child outcome variables. As well, this first model allows us to obtain a measure of the income effect.

Model 2 applies a number of controls for parental characteristics and other variables that may affect children's outcomes (independent of their effect on income) to the original bivariate regression analysis. This serves to reduce the effect of income on the developmental outcomes of children, but represents a closer approximation of the "true" effect of income.

In the final regression series we investigate whether the home environment, following the method of Blau (1999), can serve to add to our understanding of the relationship between income security and children's developmental outcomes. Blau wants to know whether purchased goods and services contribute to children's development and uses the HOME measure as a summary measure of certain aspects of the quality and quantity of parental inputs. Therefore, Blau argues, the HOME measure provides information on whether input demand is sensitive to income. Blau did find that income, particularly "permanent" income has a moderate effect on the HOME measure. We decided to test this by using the four Home environment proxy variables we created.

5.1 Descriptive Statistics for the Independent Variables and Outcomes

In Table 9 we have provided the simple descriptive data for our independent variables used in our regression analysis. The relevant means and standard deviations of each group of outcomes, are discussed below.

5.1.1 Motor and Social Development

The sample of children for this variable is almost evenly split between boys and girls, with an average of 1.2 siblings. They are most likely living in a large urban area in a two-parent family, and have not experienced any change in their family's circumstances over the period from 1994 to 1996 (Table 9). Mothers of these children had a mean age of 29 when the child was born and are likely to be working full-time. Most of the children's mothers have completed some post-secondary education or higher, and a relatively large proportion of mothers are immigrants (17 per cent). The mean household income averaged over the period 1994 to 1996, was just over $49,000.

5.1.2 The Peabody Picture Vocabulary Test

Children in this sample have a slightly higher number of siblings (1.3) than we observed with the MSD sample (Table 9). As well, there is a slightly larger proportion of girls than boys. This is likely to due to the fact that the sample is from older children who are more likely to have a number of younger siblings. Like the MSD sample, most of these children lived in large urban areas with over 100,000 people, and they had not experienced a change in family structure over the 1994 to 1996 period. The mean age of the mothers at the birth of the child is slightly higher (29.8 years) than we saw with the MSD sample. The children's mothers were still more likely to be working full-time and have received at least some post-secondary education or higher. As well, the mothers of the children were not likely to be an immigrant, although a relatively large proportion were likely to be immigrants (15.6 per cent). The mean household income over the 1994 to 1996 period was also higher in this instance ($52,000) than we found with the MSD sample.

5.1.3 Behaviour Scales

Since the behaviour scales are composed from the same age group and because the descriptive statistics are so similar, we concentrate in this section on the results for the Hyperactivity-Inattention behaviour scale. A study of the results in Table 9 show that the Hyperactivity-Inattention results are almost identical to those for the other behaviour variables. Nevertheless, we do present all of the data for the other behaviour variables in Table 9, it is just not discussed here.

Children in the sample for Hyperactivity-Inattention tended to have slightly more siblings than in the previous two samples (1.4), again likely due to the older age group observed. As well, the age of the mother at the child's birth is a bit higher, 29.8 years. However, the vast majority of these children did not experience any change in their family structure (94 per cent). As we found with the MSD there are slightly more boys than girls in this sample, but as with the PPVT there is a relatively large proportion of mothers who immigrated to Canada (16 per cent). These children are still more likely to be living in large urban areas, and to have a mother who has at least some post-secondary education and who is working full-time. The mean household income of almost $54,000, averaged over the two years, is also higher than found in the MSD and PPVT samples.

5.1.4 Math and Reading Scores

Each of these two outcomes uses the same age group sample and have almost identical results. Children in each of these samples has almost 1.5 siblings, a mean age of their mother at the child's birth of 28 years, and are likely to live in large urban areas. There is an even proportion of boys and girls. Approximately 17 per cent of the children's mothers are immigrants, with the majority of mothers having at least some post-secondary education or higher. The vast majority of mothers work full-time. And the children are not likely to have experienced any structural change to their family situation over the period 1994 to 1996. The average household income of the sample's children is approximately $56,000.

5.2 Regression Results for the Dependent Child Outcome Variables

As mentioned in our methodology section, there are two models which are used for each child outcome dependent variable. The second regression models are reduced form in nature and are compared to the baseline bivariate regression model to observe changes.

5.2.1 Motor and Social Development Scores

An inspection of the results for the bivariate relationship between each of the four income variables and motor and social development can be found in Table 10. Since we have divided the permanent or average income for the period 1994 to 1996 into units of $10,000, we can observe the effect that a one unit change in income would have in standard deviations of our dependent variable. The unstandardized regression coefficient for Average Income 1994-1996 indicates that a change of $10,000 in "permanent" household income is associated with an increase in MSD scores of approximately one-half a point (0.469). The average income coefficient is significant at the 1 per cent level. Our logarithmic income measure is observed to have a coefficient that is similarly significant and positively related to MSD scores for young children. A one unit change in the logarithmic scale is associated with an increase of almost 6 points of the MSD. This represents a magnitude of change of over 40 per cent of a standard deviation in MSD scores. Observing the coefficients for the categorized household income variable reveals that higher income categories are associated with larger increases in the MSD scale (Table 10). Our reference category for this variable is household incomes less than $20,000. Only the two highest household income category coefficients attain statistical significance. However, these two categories confer on children's MSD scores approximately 3.2 and 4.4 points, respectively, higher scores than children in the lowest income category. Children living in poor households in 1996 were likely to score almost three and one-half points lower on the MSD scale than children in non-poor households. This result is statistically significant at the 1 per cent level.

If we compare the unstandardized coefficients for these four income measures we can gain some idea of the magnitude or relative strength of the effect each income measure has on our dependent variable. The weakest effect is that of the "permanent" average income measure (0.099), then the poverty measure (-0.105), followed closely by the logarithmic income measure (0.109) and finally the highest income category ($65,000 and over) with a standardized coefficient of 0.133. Not surprisingly, having a high household income is significantly associated with higher MSD scores. These results are in keeping with the research literature where the effect of income prior to controlling for other influences is quite strong.

What is the effect on this relationship of adding in control variables? Turning to the results for Model 2 the unstandardized coefficients for the average income and poverty measures are observed to be lower than in Model 1 (0.447 and -3.326 respectively). Controlling for these parental and background characteristics reduces the initially observed effect of income on MSD scores. However, for the logarithmic and the categorized income measures the effect of the controls serves to increase the unstandardized regression coefficients. Whereas a one unit increase in logarithmic income was associated with an increase of 5.8 points on the MSD in Model 1, a similar income increase is now associated with an increase of 6.6 points on the MSD. A comparable result occurs for the categorized income category. Our unstandardized regression coefficients are ordered in the same manner as for Model 1. Children in households with incomes of $65,000 and higher are much more likely to have higher MSD scores.

These results differ from those found in the American literature (Blau, 1999; Mayer, 1997). It may be that certain of our control variables are causally connected to household income in a direct manner or that the causal relationship is difficult to ascertain temporally. In either case the variable(s) would cause the income effect to be inflated. Nevertheless, our control variables were chosen to meet those used in other such income security studies (Duncan et al., 1998; Duncan and Brooks-Gunn, 1997; Mayer, 1997).

5.2.2 The PPVT Scores

The simple bivariate effect of our four income measures on children's scores on the PPVT are statistically significant and quite considerable in some instances, as with the MSD scores. Raising household incomes by $10,000 would result in an increase, on average, of 1.2 points in children's scores on the PPVT scale, or about 25 per cent of a standard deviation. Being a child in families with household incomes of $65,000 and over, results in PPVT scores which are over 11.5 points higher than children from families with household incomes under $20,000. Being a poor child likely means a lower PPVT score, 7.3 points lower, compared to non-poor children. These results are larger than those observed with the MSD. Both of these scales are standardized with means of 100, so that income changes have a much greater impact on the cognitive development of 4 to 7 year olds than for younger children under the age of 4. The standardized coefficients show that categorized income and logarithmic income have relatively stronger effects than average income and our poverty measure.

The introduction of control variables in Model 2 serves to reduce quite considerably the effect of each of our four income variables. While all four measures remain statistically significant, their impact on children's PPVT scores is much less. Children from higher income families ($65,000 and over) are now seen to have PPVT scores that are approximately 8 points higher than children from lower income families, on average. Poor children score only 4.5 points below non-poor on average. This is still a considerable difference but rather smaller than the previous average gap of 7.3 points found in Model 1.

5.2.3 Behavioural Outcomes

We use six behaviour scales from the NSLCY to measure the social and psychological development of children aged 4 to 11. The results from our Model 1 analysis without controls indicates that children from lower income households are likely to score lower on the prosocial behaviour scale (they are less prosocial), and higher on each of the five other behavioural scales. That is, lower income children are on average more likely to be hyperactive, aggressive, experience emotional disorder-anxiety, display indirect aggression and commit property offences. A comparison of the standardized regression coefficients shows categorized household income to have the greatest effect on all six behavioural variables, followed by logarithmic household income. Income has its greatest effect on the hyperactivity-inattention, property offences and emotional disorder-anxiety scales. Its effect is least on the prosocial behaviour scale. Nevertheless, despite the fact that income has an effect on these developmental outcomes, the effects are relatively small. For example, a one unit increase of $10,000 in average household income would result in a decrease on the hyperactivity scale of approximately one-seventh of a standard deviation. We would need a $70,000 change in household income to effect a change of one point on the hyperactivity scale which has a standard deviation of 3.6 points.

Introducing controls into the regression models (Model 2) of the behaviour variables acts to decrease the effect of income noticeably (Table 10). The effect of average household income on hyperactivity-inattention and property offences, is almost halved. A small increase in the effect of income on prosocial behaviour is observed after we have controlled for parental and background characteristics. But the overall effect of this increase is extremely small, one-hundredth of a point higher than before the controls were applied for the average household income unstandardized coefficient. Categorized income and logarithmic income are still the strongest variables, when comparing the standardized coefficients. As well, hyperactivity-inattention, property offences and emotional disorder-anxiety still experience the strongest income effects, when comparing standardized coefficients.

5.2.4 Math and Reading Outcomes

Results of the math and reading tests for Children aged 7 to 11 are shown in Table 10. The table figures indicate that income has a greater effect on children's reading scores than on their math scores. A $10,000 increase in average "permanent" household income results in an increase of 0.14 points higher on the reading and 0.1 points on the math scales. Children from higher income households ($65,000 and greater) score on average almost two points (1.713) higher than children from lower income households (less than $20,000) on the reading and math variables. Categorized income and logarithmic income have the strongest effect for both reading and math, when comparing the standardized regression coefficients.

5.2.5 Home Environment Proxies

We now turn to an analysis of the extent to which income is associated with the home environment of the child. As previously mentioned, the proxy variables we have constructed do not encompass the wide range of material and information that the HOME measure used in America is able to capture. In Table 10 we observe that income is not significantly statistically related with our home environment proxy 1 for younger children (ages 2 to 3 years). However, this is not the case for the home environment proxy variables 2, 3 and 4 (for children 4 to 7, 4 to 11, and 7 to 11 respectively). Log income and our income categories reveal relatively strong effects for each of our home environment proxies 2, 3 and 4 (Table 10). Children from higher income households perform considerably better on our scale. Children aged 4 to 7 years from households with incomes equal to or greater than $65,000 score almost one point (0.870) higher on our five point scale than children from families with incomes below $20,000. This relationship is attenuated somewhat when controls are added but is still relatively large (0.676 of a point). These results are duplicated for proxy 3 and proxy 4 but to a lesser degree. Children from higher income families are relatively more likely to have home environments that are cognitively more stimulating than children from lower income homes, as measured with our proxy variables.

Blau (1999) compares the effect of income on the HOME measure to income's effect on a selected number of developmental outcomes, and determines that income has a much larger effect on the HOME measure than on any of the outcomes. We do not claim quite so much, but our results are suggestive that income certainly plays a relatively moderate role in affecting the home environment as measured here. Even with the modest scale our home environment proxy variables (2, 3 and 4) have effects which are comparable in magnitude to those observed for the behavioural variables in the regressions discussed above. The magnitude of the proxy variable effects are not however, as large as those observed for the PPVT or the Reading scores. Our analysis however does seem to support the view that income does affect input demand, as found by Blau (1999). However, this is a very tentative finding given the nature of the our home environment proxy variables and the relatively moderate to weak associations we found.

Contents

Children and Familial Economic Welfare: The Effect of Income on Child Development - April 2001