Story 7: What makes a good parent?

Story 7: What makes a good parent? Recap: Control variables Non-­‐experimental designs have particular issues in establishing cause and effect as they are not able to hold constant all factors which may be contributing to an outcome to isolate the effect of a particular ‘cause’. In a non-­‐experimental design the process of ‘fixing’ characteristics has to take place during analysis of the data, and is achieved by introducing one or more (sometimes many) control variables. The problem In deciding what makes a good parent there are a vast array of potential factors to take into account and in trying to identify the most important we must include all the other relevant factors (variables) in our analysis. As seen last week, crosstabulations are problematic when dealing with multiple control variables as you very quickly run into issues of sample size for sub-­‐groups within the tables. Therefore, we need a different analysis technique in order to allow us to take into account all the potential influences on parenting outcomes simultaneously. The analysis framework The way in which this is achieved is through one of a set of techniques (collecting known as General Linear Models) which allow us to express our outcomes (dependent variable), in this case good parenting, in terms of a combination of all the possible relevant factors (independent or control variables, sometimes also called ‘predictor’ variables). In other words our model of a good parent could be expressed like this: ‘Good parent’ = (Base value) + pinch x (Encouragement) + measure x (discipline) + large helping x (money) + moderate amount x (stability) Some evidence Levitt and Dubner (2005) in their now famous book ‘Freakonomics’ attempted to analyse what makes a good parent using this analysis framework. Using data from a US study (Early Childhood Longitudinal Study), which measured the academic progress of 20,000 children they were able to relate a very wide range of potential factors to academic achievement (their preferred measure of good parenting outcomes). The statistical technique employed here is called Regression Analysis, this technique looks at the correlations between the independent and the dependent variables in order to calculate which are the most closely related factors to specific outcomes. In this case it correlates parenting behaviours with test scores. Let’s begin with a relatively simple question: Does a child with a lot of books in their home tend to do better at school than a child with no books? The data suggests that yes there is a correlation between books in the home and test scores, the problem however is that there are also lots of other factors related to both how many books you have in the home and test scores. For example, how much money your family has is related to how many books you have, and is also related to test scores, we would therefore need to control for income in trying to establish the effect of having books in the home. Levitt and Dubner go on to study a set of 16 parental factors and identify 8 as having an influence on test scores and 8 that don’t. Here are the 16 factors, the results can be found at the end of this summary. 16 factors relating to parenting 1. Highly educated parents 2. Mother 30 or over at time of first child 3. Low birth-­‐weight 4. Read to nearly everyday 5. Mother didn’t work in early years 6. Parents speak English in the home 7. Adoption 8. Frequently watches TV 9. Parents involved in PTA 10. Many books in the home 11. Family is intact 12. High socio-­‐economic status 13. Recently moved to better neighbourhood 14. Attended Head Start 15. Regularly smacked 16. Regularly visits museums How do we measure a good parent? In this study the outcome measure (the measure of good parents) was taken to be test scores. Levitt and Dubner defend this decision in the following way: ‘Certain facets of a child’s outcome – personality, for instance or creativity – are not easily measured by data. But school performance is.’ (pg 157) This takes us back to the critical issue of measurement discussed in week 3. There are many other ‘outcomes’ of parenting that we may be interested in and we must guard against using only those that are easily measurable and treating them as if they are the only relevant outcomes. Another potential way of measuring parenting outcomes is looking at the type of person a child becomes. Flouri (2009) does this by using adult values, such as racism and authoritarianism. Before we look in detail at some of Flouri’s results, let’s reconsider our model of good parenting. ‘Good parent’ = (Base value) + pinch x (Encouragement) + measure x (discipline) + large helping x (money) + moderate amount x (stability) y = a + b1 x1 + b2 x2 + b3 x3 + b4 x4 + error
The equation in symbols sets out our working model as a regression equation. In this framework we can interpret ‘y’ as the outcome measure, ‘a’ as the value of that outcome if all the other factors are zero (so if there was no encouragement, no money etc), the ‘x’s’ are the independent variables (control variables) and the ‘b’s’ are the regression coefficients associated with each independent variable. The regression coefficients are interpreted as the impact on the outcome of a one unit increase in the associated control variable, holding all the other control variables constant. The table below illustrates how these results are usually presented for an academic audience as well as providing some guidance on how to read the table. Summary Regression analysis is a powerful tool for understanding multiple correlations between variables, where you have a clear outcome variable and a large number of potential independent variables. It is used for very many different applications across the social and medical sciences, and is perhaps the most commonly used analysis technique in quantitative social science. However, it is also working within a framework of correlations and can never conclusively establish causation, although it may help us work towards establishing causal explanations. Sources •
Levitt, S and Dubner, S (2005) Freakonomics: A Rogue Economist explores the hidden side of everything •
Flouri, E (2009) ‘Strong families, tidy houses and children’s values in adult life: Are chaotic, ‘crowded’ and ‘unstable’ homes really so bad?’ International Journal of Behavioural Development 33 (6) 496-­‐503 Answers