Using data on attitudes and behaviours to signal the presence of norms

A lack of understanding of how, where and when norms influence outcomes can result in inappropriate or ineffective interventions. This is also a risk where researchers conflate individual attitudes or behaviours with norms. However, given the lack of data on social expectations in large-scale datasets, researchers often draw on indicators of individual behaviours or attitudes to infer whether a particular norm underlies a given practice. 

In general, questions on these indicators are not enough to identify a norm because they do not necessarily shed light on collective influences. As suggested by Mackie et al. (2015), care is needed to ensure that the resulting evidence is not overstated, and further investigation should be pursued where possible. 
This section explores how existing data on attitudes and behaviours may signal the influence of gendered social norms. We first examine national household surveys, focusing on those that are cross-nationally comparable and widely available, before looking at administrative records and the possibilities presented by Big Data. 

Household surveys

Data on behaviour – for example, whether girls marry, or girls and women experience violence or can access contraception – are widely available in household surveys. These include major international survey instruments: Macro International’s Demographic Health Surveys (DHS), UNICEF’s Multiple Indicator Cluster Surveys (MICS) and the World Bank’s Living Standards Measurements Survey (LSMS). Typically, these surveys offer relatively large sample sizes (often covering tens of thousands of individuals), which allows disaggregation by sub-national area and gender, among other social and demographic characteristics. They are often repeated on a regular basis (DHS/MICS, for example, every 3-5 years), which makes it possible to chart changes over time. Pereznieto (2015) reports that such repeated cross-sectional datasets can provide insights into the effectiveness and impact of policies that affect gender norms.

Data on attitudes are also fairly accessible, to some extent in household surveys and more so in surveys of perceptions such as the World Values Survey (WVS) and regional barometers.  DHS, for example, asks women and men if they view wife beating as an acceptable punishment for a range of perceived ‘transgressions’ including burning food, leaving the house without permission and not taking care of children. The WVS, in turn, asks its respondents for their perceptions of gender roles – including gender hierarchies in education, employment, political engagement and income, the balance between work and family life, and attitudes towards gendered practices, as noted by Pereznieto (2015). For example, they are asked for their opinions on statements like ‘when jobs are scarce, men have more right to work than women’ and ‘when a woman works for pay, children suffer’. Surveys of perceptions tend to be based on smaller samples than those covered by major household surveys – often 1,000-1,500 respondents – so while gender disaggregation is possible, the resulting estimates can have relatively large margins of error.

Pereznieto (2015) also provides a useful summary of the available data on gender norms across the MICS, DHS and LSMS survey types (see figure below). 

Overview of data on gender norms from major household surveys:

Source: Pereznieto (2015)

Data on individual attitudes and behaviours can be used in tandem to shed light on whether a norm is present (see the figure below, developed by the author). Where many people report both inequitable attitudes and behaviours an uncontested norm may well be present. For example, if a high proportion of people believe that wife-beating is acceptable and reports of intimate partner violence (IPV) are relatively high – it seems likely that an uncontested norm is shaping violence-related outcomes (Quadrant III). In contrast, where attitudes are more supportive of gender equity but behaviours are less gender-equitable – for example, people hold the personal belief that wife beating is unacceptable, yet report high levels of violence – a contested norm may be influencing outcomes (Quadrant II). 

Again, these types of indicators are suggestive:  they cannot measure norms in the same way as instruments designed to disentangle collective influences from individual attitudes. However, they might be useful in the first stages of an analysis in generating hypotheses for further testing, or in situations where there is no scope for more detailed study.

Using individual-level data on attitudes and behaviours to infer the presence of norms

Mackie et al. (2015) add that high spatial or ethnic variation in a practice, especially in ‘nearby areas’, may indicate the influence of distinct reference groups that give rise to a particular norm. They also suggest that both the persistence of a practice, and conversely, a rapid shift against that practice, can signal the persistence of, or change in, a norm.

Separating community normative influences from individual outcomes

Where survey data aim to represent communities as well as individuals, the correlation of community-level variables (attitudes and behaviours) with individual outcomes has also been used to infer the presence of norms. Here, methods such as multi-level modelling permit the separation of distinct influences that occur at different spatial scales, as shown by Mackie et al. (2015). This, in turn, can help to determine whether a norm has a significant influence on behaviour and, therefore, inform strategies to change that norm. The DHS is particularly well suited to such analyses because of its large sample size and its cluster design, as clusters can be used as proxies for reference groups.

• In Nicaragua, using data from the 2011/12 DHS of nearly 3,000 female adolescents, Mendez Rojas et al. (2016) show that community norms – as proxied by a relatively young median age for sexual debut – influence the probability of rapid sexual onset and child-bearing among teenage girls, after controlling for other individual and household attributes.
• For Nigeria, Benebo et al. (2018) used data from the 2013 DHS on around 21,000 women and 17,000 men to find that community norms – as proxied by the share of men who reported that wife beating was acceptable – influenced women’s reports of intimate partner violence, and reversed the effect of women’s status, measured at an individual level.
• For Mali, Kaggwa et al. (2008) drew on data from the 2001 DHS for over 7,500 women to discover that community normative factors – proxied by collective exposure to family planning and mean births per woman – did not influence a woman’s likelihood of adopting modern contraception once other individual factors were accounted for.

One limitation of these studies, however, is their reliance on the aggregation of individual attitudes and/or behaviours to represent collective norms. Social norms theory acknowledges that some norms persist because powerful people enforce their compliance, as argued by Heise and Cislaghi (2016). The implication is that norms may reflect the preferences of the more powerful rather than an average across individuals.

Several datasets offer cross-national gendered data related to norms: the WHO Global Health Observatory, the World Bank’s Women, Business and the Law, ILOSTAT, FAO’s Gender and Land Rights database, UNICRI’s International Crime Victim Survey, and IPU’s Women in National Parliaments, among others. The World Bank’s Gender Data Portal assembles many of these indicators, grouped into several categories: demography, education, health, economic opportunities, public life and decision-making, and agency. Such systematic quantitative data has been used to provide useful insights into how norms may vary across places – even at the country level – as the Social Institutions and Gender Index (SIGI) confirms (see the box below).

Administrative data

Administrative data tend to be collected by government departments or agencies as a by-product of their routine administrative processes (such as the registration of people or the delivery of a social benefit), rather than for research or statistical purposes. As a data source, they offer some clear advantages related to cost, frequency and sample size but their coverage is incomplete in many countries and the quality of data is problematic, as noted by Alkire and Samman (2014). Even so, they provide data on norms-related practices such as education, health, births, deaths, marriages and land tenure and inheritance rights, and are increasingly seen as a useful source of information for evaluations given their large sample sizes, spatial disaggregation and frequent updates, as noted by Pereznieto (2015) (see figure below).  

Limitations of traditional data sources

Despite their ability to provide insights at scale, data from surveys and administrative sources also present some key limitations, as pointed out by Pereznieto (2015). They may be affected by social desirability bias or so-called ‘experimenter demand’ effects, with respondents misrepresenting their own views to conform to what they think the interviewer wants to hear. Other limitations include changes in how questions are framed and the resulting indicators, which may compromise comparability over time. Some hard-to-reach groups may be omitted from surveys, reducing their representativeness. Finally, standard survey questions may overlook relevant and context-specific nuances in, for example, places where girls may marry formally but may not live with their husbands until they are older.

Big data

Big data relies on so-called ‘digital exhaust’ or data generated as a by-product of people’s interaction with new technologies. It can be used to infer the potential presence of norms, offering huge advantages in terms of scale, costs and timeliness.  One common tool used in big data analytics is the identification of the range and frequency of attitudes (so-called sentiment analysis) and behaviours reported in social media such as Twitter, Facebook, Snapchat or Instagram, using computer-based coding. Mejia et al. (2018) , for example, have used Twitter data to identify changes in attitudes about sexual behaviour and gender-based violence to improve programming that aims to reduce HIV/AIDS transmission among young women in countries where its incidence is high (see box below). 

Big data approaches to norms measurement also have their own limitations, however, as confirmed by Mejia et al. (2017) and Vaitla (2018). These include potential social desirability bias – particularly given the public nature of social media; the over-representation of more affluent, urban areas; and ethical issues surrounding the confidentiality of public data.