Using behavioral insights in early childhood interventions: the effects of Crianza Positiva e-messaging program on parental investment

Abstract

We study whether an e-messaging program rooted on behavioral economics insights and administered on top of a parenting workshop helps improve parental investment and commitment. Treated families received messages thrice a week for 24 weeks. The messages were designed to help parents reorient their attention towards positive parenting goals, simplify parental tasks, and reinforce positive identities. Using an RCT on 24 early childhood centers in Uruguay, we find incremental effects over the workshop on the frequency of parental involvement and parenting quality. Effects range around 0.3 standard deviations and are larger for families experiencing more negative shocks and lower negative identity at baseline, suggesting the program triggered the right channels.

1 Introduction

The importance of a nurturing environment for child development has been well established in the psychology, neuroscience, and economics literature. Research has also shown that this environment can be substantially enhanced by programs that expand and increase the quality of pre-school care, and by interventions that foster parental competences. Although the policy agendas in many countries have increased their focus on institutional early childcare, parenting programs are yet scarce, targeted at specific populations, and in many cases short-lived and too costly to apply. The challenge is to design cost-effective parenting interventions that can be scaled up to broad fractions of the population and are able to sustain parental behaviors over time.

In this paper, we use random assignment to evaluate a text and audio e-messaging program built on top of a parenting workshop and aimed at reinforcing and sustaining positive parenting competences. We assess the effect of the program on the quantity and the quality of parental investment reported by parents. The intervention is based on behavioral economics insights, in addition to early child development science. Its design recognizes that despite parents’ good intentions, behavioral biases such as time inconsistency, cognitive fatigue and inattention, and negative identity, threaten the caregiver-child attachment, prevent parents from investing optimally, and ultimately affect child development. By focusing on ways to overcome behavioral biases, behavioral economic interventions offer a set of promising tools to improve the environments in which children grow up and develop.

The messaging program that we evaluate is one of the components of Crianza Positiva, an intervention for parents of children aged 0–2 designed to promote positive parenting practices and competences. The program delivers voice and text messages to families for six months right after these families have completed an eight-week workshop at early childhood centers. The covered topics include observing, interpreting, and responding sensitively to the child’s signals, the importance of a safe and nurturing environment, the importance of speaking and reading to the child, the key role of free play, and the value of self-caring and of having a reflective parenting attitude. The messages, delivered three times a week, aim at hel** parents refocus their attention towards positive parenting goals, decompose complex tasks into simpler ones, and reinforce positive parental identities. After workshop completion, we randomly assigned families to a treatment and a control group. Families in both groups participated in the 8-week workshop, but only those in the treatment group received text and audio messages after completing the workshop. Our analysis assesses the incremental effects of the messaging intervention over the parenting workshop a year after parents initiate the workshop and three to five months after receiving the last message.¹

The intervention makes use of mobile messaging, a low-cost tool that can help foster parental engagement and contribute to habit formation. Messages can provide continuous encouragement, support and reinforcement to parents over extended periods of time (York et al., 2019). In addition to their low cost, which makes them easily scalable, the use of mobile devices is widespread across large segments of the population, making the outreach of such programs nearly universal.²

We find that the addition of the messaging component of Crianza Positiva to the parenting workshop increased both the quantity of parental investment, as measured by the frequency of parental involvement with the child, and the quality of parental investment, given by measures of parental outreach for social support and parental reflective capacity. The effects on the frequency of parental involvement range around 0.24 standard deviations and the effects on investment quality range around 0.25 standard deviations. We also explore some potential mechanisms behind the findings. Families initially exposed to high levels of negative shocks (our proxy for cognitive fatigue) or experiencing low parental self-efficacy (our proxy for negative beliefs) show larger responses to the messaging intervention. We do not find, on the other hand, differential responses by parental time discount rates.

Our study contributes to a growing literature on early-childhood interventions in several ways. First, our results support the use of text and voice messaging in combination with behavioral economics insights as a cost-effective tool to improve child-nurturing practices. We are upfront about the fact that our study cannot establish the stand-alone effects of the text messaging program. Because the messages were nested within a broader intervention that included a parenting workshop, our results could stem from the interaction between the learning that takes place in the workshop and the subsequent messaging program. Still, we believe that at a minimum, our results provide evidence about cost-effective ways of enhancing and sustaining over time the effects of face-to-face parenting programs, which have either been shown to fade out over time (as is the case of group-based parenting workshops) or have been quite expensive to maintain (as is the case of home visits). Second, there are a few recent papers on early-childhood parenting interventions that combine nudging with e-messaging for the US. But we are unaware of other papers, aside from our own, showing similar evidence for the develo** world. Third, we go beyond previous literature by exploring the heterogeneity of parental responses when parents face different preferences, initial beliefs, and constraints. This helps us shed light on the relative strengths of different nudges. Our heterogeneity analysis suggests that nudges targeted at addressing cognitive fatigue, such as suggestions of simple actions, or nudges that boost parental self-esteem seem to be particularly effective in promoting parental investment. Finally, while most of the messaging programs we are aware of deal with parents of children that are at least in preschool, our program targets parents of children aged 0–2, an age when parental investment can have the largest returns.

The paper is structured as follows. In Section II, we provide a review of background literature, including the literature on socioeconomic gaps in early childhood investment and its relationship to child development, the recent findings of behavioral economics on parental decision-making, the use of technology in early childhood interventions, and the effectiveness of text-messaging programs. Section III provides a description of the Crianza Positiva program and of its text and audio messaging component. In section IV we describe the experimental design and evaluation instruments, assess compliance with randomization, attrition and balance, and present descriptive statistics of the data. Results are presented in section V, and we discuss and conclude in section VI.

2 Background

2.1 Socioeconomic gaps in early childhood and the importance of parental investment

Family environments in early life largely predict skill development. Heckman and co-authors underscore the importance of parental investment in the development of children’s cognitive and non-cognitive skills, not only in the short-run, but also as a determinant of long-run social and economic success (Cunha et al., 2006).

Families of low socioeconomic status spend less time with their children (Guryan et al., 2008; Schady & Berlinski, 2015b), show them fewer expressions of affection, are more likely to physically punish them (Bradley et al., 2001), and are also less likely to read to their children and to talk to them (Schady & Berlinski, 2015b; Bradley et al., 2001). The Online Appendix (Section I) offers an extensive literature that documents socioeconomic differences in parenting practices during early childhood.

2.2 Behavioral economics and parental decision-making

To a large extent, parental interventions have been designed assuming that individuals act rationally (Gennetian et al., 2016). However, many decisions can be difficult to analyze and understand through the lens of the rational model. Recent findings at the intersection of psychology and economics -behavioral economics- are changing the way we understand how individuals make decisions and behave, offering new opportunities for the design of public policies (Rabin, 1998; Thaler & Sunstein, 2009; Congdon et al., 2011). In the area of early childhood development, behavioral economics has identified the following key barriers to parental investment: a) present-bias – families with large discount rates are less likely to invest in activities that provide benefits only in the future; b) inattention and cognitive fatigue – vulnerability of poor families may reduce the idle cognitive capacity to make accurate decisions; c) Negative identities – the absence of self-esteem and self-confidence prevent parents’ motivation to undertake child rearing; d) status quo inertia – it is costly for parents to change their habits and acquire new parental skills. The Online Appendix (Section II) provides evidence about these biases in previous research and offers nudges to overcome these challenges.

Behavioral economics has grown rapidly due to its ability to explain sub-optimal outcomes and for its implications for public policies. Public policies that incorporate behavioral economics insights have the potential to be highly cost-effective once they acknowledge that small changes in the way information is transmitted, or in the way choices are presented, can have large impacts on individual behavior.

2.3 The use of technology in behavioral economics interventions in early childhood

Interventions using technology to support parents are increasingly drawing attention due to their potential of expanding programs’ reach at a low cost. Escueta et al. (2017) provide a review of the literature on interventions that use technology to support education decisions. The authors review five studies that experimentally evaluate programs that promote parental involvement in parents of preschool-aged children. They find positive effects in all the studies reviewed, which suggests that the use of technology holds great promise for early childhood interventions. In particular, recent text-messaging interventions to foster parenting investments have found favorable impacts on parents’ involvement and child literacy skills (York et al., 2019; Doss et al., 2017; Meuwissen et al., 2017; Mayer et al., 2019; Hurwitz et al., 2015). The Section III of the Online Appendix describes these interventions in detail.

2.4 The effectiveness of text-messaging programs

The effectiveness of e-messaging programs is highly dependent on their design. Cortes et al. (2018) find that parenting programs based on text messages can provide excessive or insufficient information. A three-text-per-week approach that includes information, actionable advice and encouragement is more effective to improve parenting practices than approaches that include only one text per week or that include five. Fricke et al. (2018) analyze opt out of text messaging programs that aim to improve school readiness and find that a high quantity of texts and more complex texts lead recipients to opt out more. Moreover, the authors find that programs that provide context and encouragement have lower opt out. In the context of a field experiment with charity, Damgaard and Gravert (2018) find that reminders via text messages and e-mails increase the intended behavior but also the avoidance behavior in terms of un-subscription from the mailing list.

2.5 Crianza Positiva and the text and audio messaging intervention

2.5.1 Crianza Positiva: a multi-level program aimed at strengthening parental competences

Crianza Positiva is a brief, preventive, highly protocolized and evidence-based intervention aimed at improving parenting practices and reinforcing child development. The program was originally designed to be implemented at “Children and Family Care Centers” (CAIF) of Uruguay. CAIF centers are publicly funded, privately-managed³ early childhood centers, whose purpose is “to guarantee the protection and promote the rights of children since their conception until the age of 3, prioritizing the access of those who come from families in poverty and/or social vulnerability”.⁴ We invited all CAIF centers across the country to participate in the Crianza Positiva intervention through a presentation at the CAIF National Committee. Because of limited resources, we chose to deliver the intervention to the first 24 CAIFs that expressed interest in participating.⁵ Families had no active role in enrollment decisions. They were automatically assigned to the program as long as they were attending “Experiencias Oportunas” (Timely Experiences), a weekly space at CAIF centers oriented to children aged zero to two and their caregivers.

The Crianza Positiva intervention delivered in 2017–2018 involved two modules. In the first module, families participated in a group-based workshop of eight weekly sessions, organized around four parental competences: attachment, stimulation, protection, and reflection. The workshop was delivered within the “Experiencias Oportunas” space in 2017. The second module consisted of a series of text and audio messages sent to families right after completing the workshop. This module seeks to help families incorporate the skills introduced in the workshop to their daily routines and nudge parents towards sustaining good parenting habits over time. This is the component that we evaluate in this paper.

Crianza Positiva builds on the principles of positive parenting. Positive parenting encourages the creation of sensitive and structured environments at the family level, promotes the stimulation, support and recognition of children, and trains parents to be agents of change, competent, and able to positively influence their lives and the lives of their children.⁶⁷

2.5.2 The Crianza Positiva text and audio messaging component

The messaging component of Crianza Positiva consists of 72 messages sent to families three times a week over a period of 24 weeks. We chose to send three messages per week following the finding in Cortes et al. (2018) that the three-message-per-week approach is more effective than other approaches with fewer or more messages per week. The messages were sent right after families completed the workshop. Treated families received the same messages both in text (via SMS) and in audio format (via WhatsApp).⁸ Audio messages had exactly the same content as text messages, except that the latter were personalized with the name and the gender of the child. Messages were sent on Mondays, Tuesdays and Fridays at 6 pm. Each family provided one or more mobile phones and messages were delivered to all family members willing to receive them.⁹

The 24 weeks of intervention were divided into 12 modules of two weeks. Each module referred to a topic that was discussed in the Crianza Positiva workshop. Unlike most other messaging programs that target specific goals, such as reading and speaking to the child, our program covers a comprehensive range of parenting competences. These include sensitive observation and response, gentle treatment, safety at home, incorporation of routines, speaking to the child, playing with the child, parental self-care, parental involvement, and parental reflection. Table 1 depicts the topics in the program by week of intervention.

Table 1 Topics of messages by week of intervention and associated parental competence

The message structure was designed to address behavioral biases associated with low parental investment. Before designing the messages, we conducted a preliminary analysis exploring evidence of behavioral biases in parenting behavior. The analysis collected data through a self-administered questionnaire prior to families’ participation in the Crianza Positiva workshop. Results from this analysis are presented in the Online Appendix (Section V).¹⁰ We found that the frequency of parental involvement in stimulating activities with the child correlated negatively with the parents’ time discount rate, suggesting that present-oriented caregivers place a higher weight on the current costs of parental investment relative to future benefits.¹¹ Parental investment in stimulating activities decreased also with parental stress, suggesting that inattention and cognitive fatigue could be behind sub-optimal parenting decisions (Cooklin et al., 2012). We identified a similar negative relationship between stress and parental involvement in physical games. Finally, parental involvement in physical games and social activities correlated positively with parents’ sense of competence, suggesting that parental identity and sense of self-efficacy play a role at some level on parental investment decisions.

Based on behavioral economics theory and on the exploration of its predictions using the baseline data, we built messages around the following theory of change: (i) reminding parents about the benefits of engaging in positive parenting behavior will make these benefits more salient and tangible, in particular for present-biased parents, and improve investment through activation and recall of prior knowledge; (ii) providing parents with suggestions of simple and concrete activities will help address inattention and decision fatigue by decomposing the complexity of parenting into simpler tasks; (iii) providing parents with self-care suggestions and techniques will help address inattention and decision fatigue by improving parental self-control and emotional regulation; (iv) encouraging parents to continue trying, to rely on their own resources and on the support of others, will help them overcome negative identities and strengthen parental self-efficacy.

The structure of the messages followed a 2-weeks pattern. During the first week, the Monday message sent information on the importance of a certain parenting competence. On Tuesday, parents received a suggestion to engage in a specific activity with the child in relation to that competence. On Friday, parents were invited to reflect on their performance during the week and on their personal feelings regarding the task proposed, and usually received an encouragement message. The Monday message on the second week reinforced parental self-efficacy; on Tuesday a new task on the same topic was suggested; and the last message on Friday reinforced the importance of the parental competence discussed in the 2-week module and motivation to continue practicing in the future (see Table 2 for examples of messages).

Table 2 Examples of messages by type of behavioral bias

The messages included a few other components that aimed to strengthen personal commitment and to provide parents with other sources of information and ideas. First, in the last day of the Crianza Positiva workshop, participants were asked to choose three behaviors that they could commit to practice in the following months and that they would like to be reminded of in the future. The options involved behaviors related to the topics covered in the messages. We used these selections to send each treated family a reminder of their commitments on the last day of the bi-weekly module.

In addition, treated families were provided a username and password via SMS to access “Radio Butiá”, a Uruguayan web server that hosts Latin American songs and stories online. We also directed families via SMS to access a virtual platform in Facebook to find additional information about specific topics that were mentioned in the messages. We updated this information every week. This page was mostly informative and did not encourage interaction.

3 Impact evaluation: methodology

3.1 Experimental design

3.1.1 Sample and randomization

The Crianza Positiva workshop took place between September and November 2017 at 24 CAIF early childhood centers. After workshop completion, we conducted an RCT to assess the effects of the Crianza Positiva text and audio messages component.¹² We opted for a two-stage randomization design that would enable us to assess the degree of spillovers from treated families to control families (Baird et al., 2018) in addition to the main effect of the messaging intervention. 529 families from the 24 CAIF centers were randomized to treatment in two steps (Fig. 1 illustrates the randomization process). First, we stratified early childhood centers by average maternal education¹³ and within each strata randomly allocated 60% of centers to a treatment arm and 40% of centers to a control arm. As a result of the first step of the randomization, 14 early childhood centers (296 families) were assigned to treatment and 10 centers were assigned to a pure control (none of the 233 families in these centers were treated). Secondly, within each center in the treatment arm, we randomized families into receiving or not receiving messages. In centers that were treated, 80% of families were selected to receive messages.¹⁴ From this second stage, 237 families were randomly assigned to receive messages and 59 were assigned to control.

Fig. 1

Randomization

3.1.2 Messaging intervention timeline and implementation

The messaging intervention was held between January 5, 2018 and June 27, 2018. The school year in Uruguay runs from March to December, so one third of the program was delivered during summer holidays. We asked families to provide us with at least two mobile numbers where they could be reached. When no information was available, we asked the CAIF center to provide us with the numbers. We delivered the intervention to all the contact numbers we had for each family (mothers, fathers and other caregivers). This allowed us to maximize the chances of reaching the family and to increase males’ participation (most of the participants at the Crianza Positiva workshop were women). In total, we had 373 mobile phone numbers corresponding to 237 treated families.

We sent all treated families a welcome message (both via SMS and WhatsApp) on January 5, 2018 and a closing message on June 27, 2018 that thanked them for participating in the program. Control families received a single SMS message on January 26, 2018 thanking them for participating in the Crianza Positiva workshop.

Text messages were delivered through a platform that enabled us to send the same message to all families at once. Audio messages were sent via a WhatsApp broadcast list. Due to the specific characteristics of broadcast lists in WhatsApp, these messages could only be delivered to recipients who had saved the Crianza Positiva phone number in their contact list.¹⁵

3.2 Compliance

SMS messages were sent as planned, but we could not control whether they were delivered or received. Messages could fail because of incorrect or unavailable mobile number, busy line, or no credit. We re-contacted all treated families by the end of January 2018 and randomly selected a sample of treated families in March 2018 to verify whether they were getting the messages. By January, we detected that 17% families were not receiving any SMS messages (40 out of 237 treated). Whenever we could identify that messages were failing, we asked the early childhood center to update families’ mobile numbers.¹⁶ For 23% of these families, we could update at least one line, which reduced our SMS failure rate to 13% of families. Regarding WhatsApp messages, we found that in 87% of families, at least one member of the family received the messages. We also found that families that received WhatsApp messages read on average 69% of the them. Overall, we found that 11% of families (27 out of 237) did not receive SMS nor opened WhatsApp messages. In Appendix Table 9 we provide a first stage regression where the outcome variable is an indicator for whether the family actually received messages and the independent variable is the assignment to treatment (or intention to treat) status. The results show that assignment to the messaging treatment correlates strongly with receiving the messages (the correlation is 89%) and suggests that the group of compliers is similar to that randomized to treatment.

We had an additional source of failure with messages sent to cellphones belonging to one of the carriers¹⁷ between 30 January and 20 March 2018 (36% of our sample). The carrier labeled our messages as “spam” and the messages were not delivered. However, 64% of these cellphones kept receiving WhatsApp messages during this period. Furthermore, the randomization was balanced across different carriers.

Regarding the Facebook component of the intervention, we found that 83 families (35%) signed up to the Crianza Positiva Facebook informative web. We were unable to assess which families downloaded the “Radio Butiá” stories and songs.

3.3 Evaluation scales

Barlow’s et al. (2016) systematic review about the impact of group-based parenting programs on children’s outcomes shows that positive parenting is strongly associated with children’s achievements and wellbeing. Barlow’s review identified a range of parental behaviors that are important for infant attachment security, such as parental sensitivity (De Wolff & Van_Ijzendoorn, 1997) and parent’s capacity for what has been termed “reflective function” (Slade et al., 2001). Underpinned by this significant body of research that correlates measures on parenting with benefits for children, we collected data on the quantity and quality of parental investment, and on parental knowledge about positive parenting, self-efficacy, parental stress, and time preferences through a self-administered questionnaire. We collected these outcomes at baseline (before the workshop) and in a follow-up survey administered at least two months after the messaging intervention had ended (between August and November 2018). The questionnaire had to be completed by the parent or another caregiver in the presence of the interviewer either at home or at the early childhood center (depending on family preferences) and took approximately 40 min to complete.¹⁸ A general description of the questions included in the questionnaire is provided below. For further information on the construction of each instrument please refer to the Online Appendix (Section VI).

A sociodemographic section contained questions that covered demographic characteristics of the child and the respondent (usually the child’s mother), the relationship between the respondent and the child, and household characteristics, such as household composition, maternal and paternal education, maternal and paternal employment, household assets, indicators of unsatisfied basic needs in the household, and cash transfers recipient status.

To evaluate the quantity of parental investment we considered the following dimensions. First, we inquired about the frequency of parental involvement in physical, social, caring, and didactic activities with the child. These questions were taken from Cabrera et al. (2004) and were previously used in the evaluation of Early Head Start in the United States correlating parental and children’s outcomes. Second, we included items from the Father’s Involvement subscale of the Etxadi-Gangoiti Scale (Arranz Freijo et al., 2012), which gathers information about the participation of the father in the nurturing of the child, in household tasks, and the quality of his interaction with the child. This scale was underpinned by previous research that correlates these home contextual variables with children socio-emotional development (Galende et al., 2011). Third, we asked parents about material resources at home, such as availability of books and different types of toys (role playing toys, push or pull toys, musical instruments, etc), with which we built a material resources index.

To assess investment quality, we administered a subset of items from the Positive Parenting Scale (E2P), by Gómez-Muzzio and Muñoz-Quinteros (2014). The manual classifies the parent-child relationship in four groups: (i) nurturing and attachment, (ii) safety and protection, (iii) stimulation, and (iv) parental reflective capacity. Gómez-Muzzio and Muñoz-Quinteros (2014) demonstrated the validity and reliability of E2P and offered the theory and empirical evidence that correlates this parenting scale with children’s outcomes. We also added a set of items from UNICEF MICS6 questionnaire for families of children under five, inquiring about the disciplinary methods that parents used with their child in the month before the survey.

To elicit time preferences, we administered the Kirby et al. (1999)’s Monetary Choice Questionnaire (MCQ). The instrument identifies a time discount rate for each individual that ranges from 0 to 0.249. A higher value indicates a higher preference for the present. The survey also included Abidin’s (1995) Parenting Stress Index (Short Form) (PSI/SF), an instrument for parents with children between the ages of one month and 12 years old, designed to assess stress experienced when exercising parenting activities. Parental stress impacts on children’s outcomes: Killeen and Brady (2000) find - in a health intervention for the treatment of substance abuse - that mothers who had high scores on the PSI reported more emotional and behavioral problems in their children. PSI scale is divided in three subscales (i) “Parental discomfort” which identifies the discomfort that parents experience when performing parenting duties and is derived directly from personal factors that are related to parenting (tensions or conflicts), (ii) “Dysfunctional Interaction between parent and child” which assesses whether children meet parents’ expectations and the degree of satisfaction that parents have with the child, and (iii) “Difficult Child” which identifies whether the caregiver considers the child-rearing tasks easy or difficult. The scale includes, in addition, a set of questions about stressful events that the household faced in the last 12 months. The latter answers are not considered in the overall score of the PSI/SF but are used in the analysis as exogenous sources of stress.

Parents’ perceptions about their competence as parents were gathered with the Johnston and Mash (1989) version of the Parental Sense of Competence Scale (PSOC). A considerable body of research studies the association between PSOC and children’s outcomes (e.g., Egberts et al., 2015; Slagt et al., 2012; Knoche et al., 2007). We constructed two subscales of the PSOC suggested by Menéndez et al. (2011), one related to “effectiveness” and the other one capturing “controllability” of the parenting role. The former captures whether the adult feels capable and competent to act effectively as a parent. Controllability is determined by the degree to which parents feel responsible for education situations and consequences.

Finally, we assessed parental knowledge about positive parenting by including 13 True/False items.

3.4 Attrition

Out of the 237 families randomized to receive messages (ITT = 1), 72% responded to the follow-up questionnaire (171 families). The response rate for the 292 families randomized to the control arm was 78% (see Table 3). This difference in attrition between treated and control subjects is not statistically significant at usual levels. However, if we analyze response rates for each outcome and take into account missing responses in the questionnaire, some of the outcomes show larger differences (with up to nine percentage points difference). In the next section, we assess balance in covariates across treated and control families that responded to the questionnaire and check also for balance considering the missing values in different subsets of observations. In the Online Appendix (Section VII) we address the problem of outcomes with differential attrition by calculating Lee bounds on the estimates (Lee 2009).¹⁹

Table 3 Attrition by ITT

3.5 Descriptive statistics and balance

We begin by describing evaluated families and children according to a set of sociodemographic indicators reported by the family between August and December 2017 (note that the first message was sent in January 2018). We then use this data to check for randomization balance after accounting for attrition.²⁰

Table 4 presents descriptive statistics of the respondents, the children and the households at baseline, by ITT status. Mothers are around 29 years old and children are two years old on average at the time of initiation of the messaging intervention. Eight percent of the children were born prematurely. Three out of four children live with their biological father and mother, and one out of four are still being breastfed by the time the intervention begins. On average, there is one other child in the household aside from the evaluated child and 0.2 other adults aside from the child’s parents. One out of three households faces material housing problems (problems in walls or floors, overcrowding or lack of a space to cook); only 2% lack running water, 3% lack sanitation, and 21% have no access to at least one basic comfort asset, including heating, a fridge, and a water heater. We construct an asset index including 18 household and family assets and utilities.²¹ The index ranges from 0 to 0.77 with higher values indicating higher availability of assets. The average value of the asset index for families participating in the study is 0.25 with a standard deviation of 0.14. Sixty seven percent of families are recipients of government cash transfers. Eighty percent of families attended six or more sessions of the eight-session workshop. Mothers are by far the main respondents to the questionnaire (93%), followed by fathers, grandmothers and other caregivers. The child under study is the mother’s only child in 38% of the cases. Almost 30% of mothers are high school graduates; one out of three completed middle school but not high school, and the rest did not complete middle school. Fathers work full time in 79% of the cases. Nearly three out of five families report having experienced a negative shock in the past year (a death, a divorce, unemployment, money problems, problems with the law or with drugs in the family); the average number of problems is 1.34. On average, 38% of families are classified as having a high discount rate, meaning that the discount rate of the respondent is higher than 0.1 (discount rates range from 0 to 0.249). Lastly, around half of respondents have a low parental sense of competence at baseline, which implies scoring four or less in the subscale of efficacy of the Parental Sense of Competence Scale.

Table 4 Descriptive statistics and balance in covariates across treatment arms

The last two columns in Table 4 show the differences in covariates at baseline between treatment and control subjects that responded to the follow-up questionnaire. Out of 23 covariates analyzed, only mother’s age is statistically different at 1% across treatment and control subjects (mothers in the treatment group are almost two years older than mothers in the control group). Two other variables are statistically different at 10%, the number of other children in the household (which is larger for treated families), and the number of other adults in the household (which is smaller for treated families). Although we adjust for these three covariates in the regression analysis, we are confident about the balancing properties of the sample: the number of statistically significant covariates is around what we would expect to get by chance.

Table 5 shows descriptive statistics of outcomes assessed at follow-up. Each family of outcomes is identified by a heading in italics. The parental investment subscales indicate the frequency with which parents engage in different parenting activities, with one indicating “Never” and six “All or most days of the week”.²² The Parental Time Investment Index is a summation of the physical games, didactic and social activities scales: it averages 13 and has a maximum of 18. Father’s involvement in childrearing is a continuous index from zero to one with an average of 0.77. 84% of families have at least five children’s books in the household.

Table 5 Descriptive statistics of outcomes at follow-up

The Positive Parenting Scale (E2P) shows values above four (on a maximum scale of five) for attachment, and values above 3.5 for routines, social support and parental reflection.²³ Parents report higher levels of stress when asked about personal discomfort and child characteristics, than when assessing the interaction with the child. The average sense of parental competence is four, on a maximum of six. The percentage of families reporting the use of some type of violence as a “disciplinary” approach was 39%.

Parenting knowledge is the summation of different true-false items indicating knowledge of positive parenting competences. On average, parents had 11 out of 13 questions right. Finally, the time discount rate (the rate at which parents discount the future) averages 0.08 in the sample, with a minimum of 0 and a maximum of 0.249.

3.6 Estimation approach

Our simplest specification uses OLS to regress Y_ic, the outcome of interest corresponding to family i in CAIF center c, on ITT_ic, an indicator taking the value of one if the family attended a CAIF center randomly assigned to receive messages, and zero otherwise; and on Spill_ic, an indicator equal to one if the family was assigned not to receive messages in a CAIF center selected to participate in the messaging intervention, and zero otherwise:

$$Y_{ic} = \beta _1ITT_{ic} + \beta _2Spill_{ic} + \varepsilon _{ic}$$

(1)

This specification accounts for the two-stage randomization structure, which was conducted to assess the degree of spillovers of the messaging intervention on CAIF families that were not receiving messages, but that attended an early childhood center in which other families were receiving messages.

We adjusted Eq. (1) subsequently for the following set of covariates: (i) Strata_c, the variable used to stratify the randomization at the CAIF center level (a dummy equal to one if the average level of education of mothers participating in Crianza Positiva at the center was at least middle school); (ii) B_ic, a set of covariates that were unbalanced after attrition at a significant level below 10% (the mother’s age, whether the child lived with other adults aside from mother and/or father, the number of other children in the household); and (iii) X_ic, other covariates at the family level that could help improve the precision of the estimation (the child’s age and gender, mother’s education, whether the child lived with both biological parents, the time elapsed between the launching of the messaging intervention and the administration of the follow-up survey, whether the family had a negative shock in the 12 months prior to the intervention, and a baseline measure of the outcome, when available). Our final specification was as follows:

$$Y_{ic} = \beta _1ITT_{ic} + \beta _2Spill_{ic} + \beta _3Strata_c + B_{ic}^\prime \beta _4 + X_{ic}^\prime \beta _5 + \varepsilon _{ic}$$

(2)

Because our experimental design involved a clustered randomization, we adjusted standard errors for the clustered design (Abadie et al., 2017). The usual approach when the number of clusters is large is to estimate standard errors using the Zeger and Liang (1986) covariance estimator. Unfortunately, the number of clusters in our data is only 24. As recommended by Cameron et al. (2008) we use a wild bootstrap with null imposed to enable more accurate cluster-robust inference. We conducted randomization inference for 2000 replications. We report the associated p values.

For each family of outcomes, we adjust standard errors to account for multiple hypotheses testing using the Romano-Wolf correction (Clarke et al., 2020). We exclude summary indices from these adjustments.

We also report the power of the sample to detect an effect of 0.25 standard deviations in the case of continuous outcomes and an effect of 10 percentage points in the case of discrete outcomes. The power calculations account for the experimental design (randomization at two levels) and for intra-cluster (intra-CAIF center) correlation of the outcome.

4 Impact evaluation: results

4.1 Core results

In the main text, we analyze the results on families that were randomized to receive messages and focus on spillovers on Appendix Table 11. Tables 6 and 7 report the results of regressions of each outcome on an Intention to Treat (ITT) indicator and an indicator for whether the subject belongs to the spillover sample (Spill). In Table 6, we report results on parental quantity of investment with respect to time and material resources and on the quality of this investment. In Table 7, we report results on parental stress, sense of parental competence, parental knowledge and discount rate.

Table 6 Intention to treat (ITT) effects (in standard deviations) on parental Investment

Table 7 Intention to Treat (ITT) effects (in standard deviations) on parental stress, sense of competence, discount rate & knowledge

The first column in each of these tables shows the raw treatment-control differences in standard deviations and the coefficient’s unadjusted standard error. The second column shows ITT effects after adjusting for the stratum used for randomizing CAIF centers (average maternal education above middle school), child’s gender and age in months, and maternal education. Column (3) adds the following covariates to the former regression: mother’s age, time elapsed since the messaging intervention begun, whether the family had a negative shock in the 12 months prior to the intervention, number of other children in the household, whether the child lived with other adults in addition to father and mother, and whether the child lived with her biological father and mother (intact family). Column (4) adds a control for the outcome at baseline, if available.²⁴ While results are quite robust across specifications, we center our discussion on that in Column (3), which improves precision and adjusts for the few imbalances in covariates across treatment arms. In columns (5) and (6) we report the p values corresponding to the outcome difference by ITT status reported in Column (3) when adjusting for clustering and multiple hypothesis testing (MHT) respectively. The seventh column reports the p value that results from randomization inference.

We also report in Column (8) the sample power to detect pre-established effect sizes. In particular, the power calculations consider an effect size of 0.25 standard deviations in the case of continuous outcomes and of 10 percentage points in the case of dichotomous outcomes. The calculations account for the experimental design (randomization at two levels) and for intra-cluster (intra-CAIF center) correlation of the outcome.

4.1.1 Parental quantity of investment: time and material resources

The first outcome in Table 6 is the aggregate index of parental time investment, which shows an ITT effect of 0.33 standard deviations. To put effects into context, Attanasio et al. (2018) evaluate a parenting intervention in Colombia and find that an increase of 0.34 standard deviations in the home environment quality could work as suggestive mechanism for an increase of 0.15 standard deviations on overall child development.²⁵ Our effect is statistically significant when adjusted by the clustered design, by multiple hypotheses testing (MHT), and by randomization inference. We also find that the intervention increases the average frequency of parental engagement in didactic activities with the child by 0.29 standard deviations. The effect is significant at a 1% level when considering unadjusted p values, and at the 5% level when adjusting for clustering, randomization inference, or MHT. We find that the messaging intervention increases the frequency of parental involvement in physical games with the child in a magnitude of 0.27 standard deviations. This estimate is significant at a level of 5% when considering unadjusted p values, clustering, or when considering p values adjusted for MHT, and at the 10% level when conducting randomization inference. The ITT effect on the frequency of parental involvement in social activities with the child has a magnitude of 0.27 standard deviations and is significant at the 5% level, but loses significance in the absence of controls (Column (1)). However, we find statistically significant effects in the frequency of involvement in social activities, didactic activities and physical games when we dichotomize the outcome and place a value of one to daily involvement in that activity and zero otherwise (See Appendix Table 10). We find no statistically significant ITT effects on father’s involvement in childrearing nor on availability of books in the household. The power to detect an effect of 0.25 standard deviations is 0.73 for the parental investment index, 0.53 for the measure of engagement in physical games, and 0.61 and 0.67 for engagement in didactic activities and social activities respectively.

4.1.2 Parental investment quality

Regarding the quality of parental investment, we find a positive effect of 0.31 standard deviations on the index of positive parenting when considering unadjusted p values. The effect maintains statistical significance at the 1% level when we run randomization inference or when we account for the clustered sample design, and at the 5% level once we account for multiple hypotheses testing. Similarly, we find that the intervention has an effect on parental outreach for social support of 0.25 standard deviations that is significant at the 5% level when considering unadjusted p values, clustering, and at the 10% level after adjusting for MHT. Moreover, we find a statistically significant ITT effect (at the 1% level when considering unadjusted p values, clustering or randomization inference, and at the 5% level after adjusting for MHT) on parents’ reflective capacity. The magnitude of the effect is of 0.33 standard deviations. Power ranges around 0.5 for all outcomes.

4.1.3 Parental stress, sense of competence, discount rate and knowledge

In Table 7 we report the effects of the messaging intervention on parental stress, sense of competence, discount rate, and knowledge about positive parenting. We find no evidence of an effect of the intervention on parental stress, discount rate, or parental knowledge. On the other hand, we find a positive and statistically significant effect on the sense of parental effectiveness when considering unadjusted p values, but the effect becomes non-significant once we adjust for MHT, and when adjusting standard errors for the clustering in the sample. The power to detect a 0.25 standard deviation effect size is 52%. The power is lower in the case of the controllability outcomes and parental knowledge. In the latter outcomes, we do not find any statistically significant effects.

4.2 Heterogeneity and mechanisms

To assess whether the messaging program operates through behavioral channels, we explore program heterogeneity across three dimensions of parental preferences, constraints, and beliefs: the parental discount rate, negative shocks faced by the household in the previous 12 months, and parental sense of competence.²⁶ Our baseline assessment suggested that behavioral barriers would be stronger in parents with high discount rates, families with more negative shocks in the past, and parents with low sense of competence. Because the intervention was designed to address these barriers, we expected it to have stronger effects among these families.

Results are presented in Table 8. Each pair of columns shows the coefficients and standard errors from an OLS regression of the outcome in each row on the ITT main effect, the interaction between ITT status and the variable capturing the behavioral barrier, the main effect of the behavioral barrier, maternal education, government assistance, and randomization strata. The table displays only the ITT main effect and the interaction of ITT with the behavioral barrier, presented in standard deviations. We adjust our results for multiple hypothesis testing using the Romano-Wolf correction (Clarke et al., 2020) and report adjusted p values in squared brackets below significant interaction coefficients. Columns (1) and (2) show that families with higher cognitive fatigue (higher likelihood of at least two negative shocks in the past 12 months)²⁷ derive some additional benefits from the intervention.²⁸ In particular, families facing a larger number of negative shocks are more likely to implement routines, are less likely to use violent discipline, and have a stronger sense of parental competence at follow-up. These results, with the exception of the likelihood of implementing routines, are robust to adjusting for multiple hypothesis testing.

Table 8 Heterogeneity by behavioral barriers. OLS regressions

Columns (3) and (4) explore whether parents with lower initial parental sense of competence benefit more from an intervention geared towards providing encouragement and constructing positive identities.²⁹ Parents with low initial self-esteem are more likely to increase the number of toys at follow-up and improve their scores in an overall score of quality of investment. The likelihood of establishing routines and reflecting on parenthood are higher for parents with low initial self-esteem when considering unadjusted p values but become non-significant when adjusting for MHT.

Results in columns (5) and (6), on the other hand, show no evidence that the intervention was more effective among parents experiencing higher discount rates.³⁰,³¹

Finally, we run an additional heterogeneity analysis considering workshop attendance to assess potential complementarities between the workshop and the intervention. We present results in Appendix Table 11. Overall, we find no evidence of stronger results among survey respondents that attended six or more sessions of the workshop.³² There is only one interaction coefficient that remains significant after adjusting for multiple hypothesis testing, but it runs in the opposite direction than hypothesized: the messaging intervention increases the discount rate more for families that attended more than six workshop sessions.

In sum, we find evidence that the messaging program has stronger effects on families with higher exposure to cognitive fatigue and more negative identities. These differential effects operate mainly on the qualitative margin. By suggesting simple activities to carry out at home, parents are more able to establish routines, need to rely less on the use of violent discipline and their sense of parental competence increases. The encouragement provided to parents through the messages seems to improve their reflection capacity, their ability to organize their routines, and material investments.

5 Conclusions

This paper evaluates the impact of a text and audio messaging program (Crianza Positiva) aimed at hel** parents develop and sustain parenting competences over time. The program, which is built on top of a parenting workshop, reminds parents about the benefits of engaging in positive parenting practices, provides them with suggestions of simple and concrete positive parenting activities, reinforces positive parental identities, and encourages parents to seek resources within their families and community to improve their parenting behaviors and attitudes.

Our paper is among the first ones, together with York et al. (2019), to show the benefits of a program combining e-messaging and nudges in boosting parental investment; and the first we are aware of that implements these tools to address parenting in a develo** country. Unlike other parenting programs using e-messages, our program covers a comprehensive range of parenting areas, including sensitive observation and response, the importance of a safe and nurturing environment, the importance of speaking and reading to the child, the key role of free play, and the value of self-caring and of having a reflective parenting attitude. Moreover, our program is focused on parents with children aged 0–2, while most of the literature focuses on parents of older children. Also, our evaluation includes an extensive set of outcomes which are measured using validated instruments. Unlike previous literature, we not only assess the quantity but also the quality of parental investment, including parental stress, parental sense of competence, parental knowledge about parenting, and parental sensitivity. The program complemented a prior 8-week workshop for families at local early childhood centers and sent caregivers text and voice messages through their cellphones (via SMS and WhatsApp, respectively). Families received both types of messages three times a week during 24 weeks between January and June 2018.

The program was well-received by families. Among families assigned to treatment, 95% said that the messages had been either very useful (61%) or somehow useful (34%), and only one family opted out of the audio messages. Furthermore, the program had a positive effect on the frequency of parental involvement with the child, on parental competences and parenting attitudes. Our findings show that messages had an incremental impact over the workshop in the range of 0.30 standard deviations on a parental time investment index and on parental engagement in social, physical, and didactic activities with the child. They also increased parents’ quality of investment as measured by a positive parenting index and by an index of outreach for social support (by 0.31 and 0.25 standard deviations respectively). Moreover, the addition of messages to the workshop improved parental capacity to reflect on parenting by 0.33 standard deviations.

We find that the program had stronger effects over parents with initial negative identity (i.e., a low sense of parental competence) and over parents experiencing negative shocks in the months prior to the intervention (higher cognitive fatigue). Because the intervention targeted these behavioral barriers, our results suggest that it may have triggered the right channels. We could also infer from these findings that nudges involving recommendations of simple actions and messages of encouragement may be particularly important to activate changes in parental decision-making.

Overall, our results indicate that the combination of e-messages and nudges are a promising tool to enhance parental behaviors, competences and attitudes. Our findings indicate that the program is highly cost-effective: a back-of-the-envelope estimate of the cost per family of implementing this program is US$ 10 if SMS were to be used, and much less if only WhatsApp messages were to be considered. In terms of external validity, our intervention was implemented in CAIF centers in Uruguay, which tend to assist families of lower socioeconomic status, but the program was designed for any socioeconomic setting and would need little adaptation to be delivered in other contexts. Moreover, the intervention was embedded directly in a governmental-provided program, which made the implementation and results close to a “real-life” intervention.

While the self-reported nature of our outcomes could be pointed out as a caveat, we don’t believe our findings are being led by social desirability bias. All parents in the treatment and control groups participated in an 8-week parenting workshop prior to the implementation of the messaging intervention, focused on the key competences associated with positive parenting. Indeed, we find no impact of the messaging intervention on an index of parenting knowledge about positive parenting competences. Furthermore, our endline measurements include mostly internationally validated scales which have been used in multiple and diverse settings, reducing concerns about the messaging program being designed to “teach to the test”.

The fact that the Crianza Positiva messaging program was implemented right after families had finished the parenting workshop raises another issue. We cannot affirm that the messages would have been effective in the absence of the workshop. However, the literature shows that short group-based parenting interventions have a hard time consolidating and sustaining effects over time, and home visits are hard to maintain due to their high costs. At a minimum, our study underscores the value of messages in hel** integrate and put into practice concepts acquired in prior face-to-face parenting interventions. This is supported by findings in Balsa et al. (2020), that show that the workshop by itself does not increase the frequency of parental involvement with the child. Under a less stringent perspective, many RCTs are conducted on top of interventions that already exist, and usually results are hard to disentangle completely from the root program.³³As such, our study offers suggestive evidence about the potential that e-messaging programs could have as stand-alone. In future research, we plan to explore the effectiveness of messages without the requirement of participating in a previous workshop. We also plan to assess whether families’ responses to the messages vary according to the sender (sender versus content effect), and to improve our understanding of the mechanisms behind the effects by randomly assigning message types.

Appendices

Appendix 1: First stage regression

Table 9

Table 9 Regression of an indicator of whether the family received any message on ITT status

Appendix 2: Intention to Treat (ITT) effects on every-day parental time investment

Table 10

Table 10 Intention to Treat (ITT) effects in standard deviations on every-day parental involvement (dichotomous indicator)

Appendix 3: Heterogeneous effects on workshop attendance

Table 11

Table 11 Heterogeneity by workshop attendance of respondent. OLS regressions

Appendix 4: Spillover effects

After participating in the workshop, parents could potentially interact in meetings, local sites, or social media. To assess spillovers, we compared the effects of non-treated families within treated centers with non-treated families within non-treated centers.

Appendix Table 12 shows the coefficients on an indicator equal to one if the family was assigned not to receive messages in a CAIF center assigned to treatment, and zero if the family belonged to a CAIF center assigned to control (β₂ in Eq. (2) when including the following controls: the child’s age and gender, mother’s age and education and whether the child lived with both biological parents). Any positive effect of the intervention on this indicator would suggest spillover effects. Each row corresponds to a different outcome. Overall, the evidence is not supportive of the idea of positive spillovers. There is only two statistically significant effects that are also robust to familywise multiple hypothesis testing adjustment, but the statistical power is quite low.³⁴ Moreover, the effects also run in the opposite way than hypothesized. Families in this group show higher levels of violent discipline and lower levels of knowledge of positive parenting competences. The results could express frustration among the untreated parents in treated centers for excluding them from participating.

Table 12 Coefficient on a dummy indicating that the family was assigned to the control group in a center assigned to treatment