Is Less More? Number of Siblings and Frequency of Maternal Activities with Preschool Children

ABSTRACT

We use the German Socio-Economic Panel to investigate to what extent the number of siblings has on the frequency of cognitively stimulating activities mothers undertake with their children aged 2 to 3 and 5 to 6. The resource dilution model assumes that this relationship should be negative. We introduce the resource augmentation hypothesis and argue that increasing sibship size can also have a positive effect when efficiency gains, the public good character of maternal activities, reallocation of maternal time, and child care by older siblings are considered. Our results from intraindividual and between family analyses suggest that resource dilution is unavoidable in large families, but that resource-augmenting processes are also at work. The relationship between the number of siblings and activity frequency is not linear and can be reversed from negative to positive in smaller families depending on the presence of younger siblings.

KEYWORDS:

• family themes
• resources
• siblings
• time

Notes

See Eschelbach (2009), Bauer and Gang (2000), or Blaess (2005) for analyses conducted in the German context.

See also Leibowitz (1977).

Additional activities suggested to improve children’s skills are those in which parents do not actively take part but make it clear that they are accessible to the child at any time and those in which parents take more responsibility in interactions with their children (see Zick et al., 2001, and the references cited therein).

In their investigation to determine how the amount of time a mother spends at work is related to the amount of time she spends in different activities with her children, Baydar, Greek, and Gritz (1999) found that increases in working hours have different effects on different child care activities but that the overall effect on all activities is negative.

Mokken scale analysis is an Item Response Theory model that allows testing the scalability of categorical items on a cumulative scale (Mokken, 1971; Molenaar, 1997). The model is probabilistic and nonparametric and can be understood as a nondeterministic extension of Guttman scaling. It poses the existence of a unidimensional latent trait related to observed items in a hierarchical way. Popular item scores or “easy” items are associated with lower scale values, whereas uncommon or “difficult” item scores are related to higher scale values. Thus, for example, if all mothers read to their children daily but only a few sing songs with them, reading to children alone would be an indicator of low activity frequencies, whereas reading and singing songs would denote high frequencies. Scale reliability is defined as a consistent progression from easy to difficult items associated with nondecreasing probabilities on a latent dimension (monotonicity) rather than a linear correlation among equally distributed items as in factor analysis. This implies that two items, an easy item and a difficult item, may belong to the same cumulative scale even if their correlation coefficient is low.

By including individual level weights in the model, the maximum likelihood iterative process does not reach convergence and parameter estimates are not reliable. Observations were therefore not weighted. To make sure that this decision does not affect our conclusions, we took advantage of the fact that random effects estimates can be approximately obtained through pooled OLS methods with clustered robust random error estimates and fitted the model using this alternative method. Results show that pooled OLS estimates of M1 with weighted data are comparable with and are for some parameters even identical to estimates for unweighted data. For this reason it is safe to assume that reported estimates on Table 2 using maximum likelihood with unweighted data are approximately the same as those we would have obtained with weighted data. To keep this result comparable with estimates from M2, results reported for fixed effects models do not include weights. On the other hand, we tested the adequacy of the random-effects assumption using a Hausman test. The test indicates that estimates might be inconsistent. To further explore this issue, we fitted models using the person-specific means (between effects) as well as deviations from means (within effects) of all time-changing variables as suggested by Allison (2009, pp. 23–25) and found out that discrepancies are limited to the parameters for household income and child’s health. We then ran the original model controlling for person-specific means for these two variables to avoid cluster level confounding and obtained comparable results for our variable of interest, the number of children. A Hausman test for this modified model suggests that the random-effects assumption of no correlation of unit-specific effects with covariates is appropriate.

There is ample evidence that higher educated mothers not only spend more time with their children than less educated mothers (Guryan, Hurst, & Schettini Kearney, 2008; Sayer et al., 2004), but they do so in qualitatively different activities, such as reading instead of watching TV (Bianchi & Robinson, 1997; Hofferth & Sandberg, 2001). Only working time, especially of women, does not appear to have a large impact on time spent with children. It has long been documented that employed mothers somehow manage to compensate for their working time by spending more time with children during nonworking hours, including weekends (Booth, Clarke-Stewart, Vandell, McCartney, & Owen, 2002; Nock & Kingston, 1988).

Household income correlates positively with more time spent by parents with their children (Guryan et al., 2008; Hill & Stafford, 1974; Zick & Bryant, 1996).

We thank the anonymous reviewer for this idea.