Uses machine learning to study parenting styles

How should we raise our children? Studies have shown that the amount of time parents spend is not the only important factor in developing children’s skills (Del Boca et al. 2014, Attanasio et al. 2016); Parenting style is also important (Fiorini and Keane 2014). Parenting style is a strategic choice associated with incentives (Doepke and Zilibotti 2014). To study the relationship between parenting style and child development, researchers relied on ad hoc perceptions or previous research to limit parenting complications to specific core tasks. For example, reading to children has been shown to be highly predictable in developing children’s skills (Kalb and Jan van Ours 2013). But how can we measure parenting style without relying on previous beliefs?

In a recent study (Rauh and Renée 2022), we adopted a method that allows data to speak. We take a model from computational linguistics to make sure what kind of parenting style exists and is most significant. The main purpose of machine learning algorithms is to learn from the co-occurrence of words and to form the subjects around them. In the context of parenting style, the idea is that parents who engage in one activity may be more likely to engage in another given activity as algorithms learn from the co-occurrence of parental actions to define two or more parenting styles.


We rely on the Quebec Longitudinal Study of Child Development (QLSCD), a detailed panel of family representative samples from Quebec, a province in Canada, where a child was born between October 1997 and July 1998. We collect information on the behavior of mothers. The target children were ages 5, 17, and 29 months when conducted across three waves.

One advantage of the dataset is that mothers’ behavior towards their children is not captured through self-reported survey questions, but through enumerators. At the end of each interview, the enumerator records ten variables and classifies whether the mother is involved in specific actions during the interview. This makes the data less sensitive to a bias that may result from mothers misreporting their behavior to their children.

In Table 1, we look at a list of ten jobs and what percentage of mothers are involved with them during the three interviews. Two things stand out. First, parents as a whole are more likely to support or support the child by kissing and hugging rather than shouting or annoying the child. Second, despite this imbalance appearing in all three study waves, there are significant changes toward more punitive and less supportive actions as the child ages.

Table 1 Parental actions across the survey wave

Comments: The table describes the respondents’ behavior and their interactions with their children during the annual QLSCD interview in percentage terms. Behaviors are assessed by the enumerator during the interview.

Parenting style

The model we use to classify parenting styles is Latent Derichlet Allocation (LDA) by Blaine et al. (2003) For classification of text in subjects. In recent work, economists have used the LDA to determine major CEO behavior (Bandiera et al. 2020) and political ideology (Draca and Schwarz 2021). In our case, each parenting style is defined by the action potential of each parent, and parents are classified into two styles. Classification is not ‘strict’ ​​in the sense that parents can be a mixture of two types: Algorithms assign a share of each style to the parent in the dataset.

In Figure 1, we show the distribution as a result of parental activity when classifying parenting into two styles, A and B. The red bars indicate a given verb in parenting style A and the introduction of blue bars in parenting style B. The left panel displays the probability of each of the ten actions in a subject. A parenting style should be imagined as a pitcher from which a parent draws action. A mother following Parenting Style A closes her eyes and draws an action from the red bar. He will probably draw the biggest bar; For example, she may give her child an educational toy or respond to her baby’s words. A parent in the parenting style will probably not draw any action when drawing from the blue bars on the belly because they are too small. In other words, parenting style B is characterized by inactivity. Based on this distribution of action, we follow developmental psychologists McCabe and Martin (1983) and label parenting style A as ‘warm’ and parenting style B as ‘cold’. On the right panel of Figure 1, we see the standard relative amplitude of a given action within a parenting style. We find that warm mothers are less likely to scold or yell at their child. Cold parents, if anything, just check on their child.

Figure 1 Distribute actions by parenting style

Comments: The left panel describes the sharing of activities for each of the two parenting styles. The right panel shows the standard importance of an action in a style by setting the average deviation to 0 and the standard deviation to 1.

Are all parents equally likely to follow a warm parenting style? To answer this question, we look at warm parenting and the relationship between parental and family characteristics. In Figure 2, we see the distribution of parental style through maternal education. On the left panel, we can see that among the less educated mothers who have the most high school education, the style of cold parenting is quite promising. This is indicated by the mass on the left side of the graph. If we look at the panel on the right, we see the opposite of mothers with college degrees. For highly educated mothers, there is a shift in the distribution to the right; They are more likely to follow a warm parenting style. We also find that young mothers and those who have more children are more likely to engage in cold parenting style.

Figure 2 Style distribution by maternal education averages across waves

Comments: Transparent bars represent the binned possibility of engaging with a warmer rather than a cold parenting style, while the solid line is the density of the kernel. Samples are pooled samples where each parent appears three times.

Table 1 indicates that the style changes over time. While there is some perseverance – that is, warm style mothers are more likely to follow a warm style and those with a cold style are more likely to continue the cold style – there are systematic changes as the baby ages. On average, the parenting style gets cold. Another significant change is that while there is no difference in parenting style between boys and girls at 5 months of age, when the child reaches 29 months of age, boys are statistically more likely to experience a cold parenting style.

Does parenting style affect skill development?

Although we cannot give a definitive answer to this question due to the lack of external diversity, we can see whether children who come in contact with a particular parenting style acquire higher skills at a later age. More specifically, we see summary measurements of cognitive test scores (such as math or reasoning) and non-cognitive scores (such as behavioral disorders or hyperactivity) at age 6.

In Figure 3, we show the effect of warm parenting on the measurement of standard skills at age 6. We show the relationship of the parenting style measured individually (above) at each age and for an overall measurement of the style calculated across the three ages (below). . We find that at 5 months of age, children who are exposed to a completely warm parenting style, instead of a completely cold parenting style, acquire cognitive skills (left) which is 0.3 standard deviation and non-cognitive skills (right) which is more than 0.2 standard deviation. . The size of the effect of parenting style at 5 months of age is greater than between 17 and 29 months of age although the parental style is measured farthest from the results when the child is 6 years of age. This provides some support for the notion that very early childhood investments are particularly important.

Figure 3 Coefficient from the return of warm parents on cognitive and non-cognitive skills at the age of 6 years

Comments: The dependent variable is calculated at the age of 6, taking the first factor of the six measurements of cognitive and non-cognitive ability each. The score is standardized with an average of zero and a standard deviation of one. Thin lines represent a 90% confidence interval.

The overall measurement of the parenting style shows a much higher correlation with the result. More than half of the standard deviation of warm guardianship is associated with higher cognitive skills and one-third of the standard deviation is associated with higher non-cognitive skills.

Look ahead

Given the indexable growth of available data, we should use machine learning to better understand what kind of parenting ‘works’. Allowing data to speak allows us to challenge traditional assumptions and uncover patterns of success. It can help policymakers and researchers design intervention and support programs to help parents navigate child-rearing complications and avoid ‘parental traps’ (Hilger 2022).


Attanasio, O, S Cattan and S Krutikova (2016), “Early Childhood Development Policy: Evidence and Research Agenda”,, 09 June.

Bandiera, O, A Pratt, A Hansen and R Sadun (2020), “CEO Behavior and Strong Performance”, Journal of Political Economy 128 (4): 1325-1369.

Blei, DM, AY Ng and MI Jordan (2003), “Assignment of Latent Derichlet”, Journal of Machine Learning Research 3: 993–1022.

Dale Boca, D, C Flynn and M Wiswall (2014), “Household Choices and Child Development”, Review of Economic Studies 81 (1): 137-185.

Draca, M and C Schwartz (2021), “How polarized are citizens? Measuring ideology from ground-up ”, SSRN, 11 May.

Doepke, M and F Zilibotti (2014), “Tiger Moms and Helicopter Parents: The Economics of Parenting Style”,, 11 October.

Fiorini, M and MP Keane (2014), “How Children’s Time Allocation Affects Cognitive and Cognitive Development”, Journal of Labor Economics 32 (4): 787–836.

Hillger, NG (2022), Parental Trap: How to Stop Parental Overloading and Resolve Our Inequality CrisisMIT Press.

Kalb, G and J van Ours (2013), “Reading to Children: A Major Beginning in Life”,, 10 June.

McCabe, E. and J. Martin (1983), “Socialization in a Family Context: Parent-Child Interaction”, Handbook of Child Psychology 4: 1-101.

Rauh, C and L Renee (2022), “How to measure parenting style?”, CEPR Leaflet 17326.

Leave a Reply

Your email address will not be published.