ISSN (Print) - 0012-9976 | ISSN (Online) - 2349-8846
-A A +A

Discriminatory Attitudes towards the Female Mobility

Evidence from a Survey Experiment

Michael Dickerson (michael_dickerson@brown.edu) is a visiting fellow at Pahle India Foundation, New Delhi.

Empirical evidence from a survey experiment administered in Bihar, Punjab, Kerala, and Delhi reveals discriminatory attitudes towards the freedom of movement of females. The findings reveal the main effect—an indication of discrimination against females in the context of mobility—to be significant and the magnitude of the main effect to vary significantly by region, though the treatment condition is still significant in each.

The author thanks Rose McDermott for her help in developing this article.
 

The empirical study of discrimination against women within India has increased substantially in recent decades, with notable contributions from Amartya Sen (1987, 1990), Jean Drèze and Geeta Kingdon (2001), Murthi et al (1995), Shankar Subramanian and Angus Deaton (1991), and Jagdish Bhagwati (1973), among others. Anand and Sen (1995) noted the many forms of gender disparities, providing a conceptual overview of various categories of inequality, including household inequality. Deaton (1997) noted the numerous challenges associated with the empirical study of intra-household phenomena, including discrimination against females. In this article, I present empirical evidence of discriminatory attitudes towards the freedom of movement of females from a survey experiment administered in Bihar, Punjab, Kerala, and Delhi (N=5,315). My findings indicate the main effect—an indication of discrimination against females in the context of mobility—to be significant, and the magnitude of the main effect to vary significantly by

region, though the treatment condition is still significant in each. Moreover, I find evidence of treatment effect heterogeneity associated with both gender and level of education of subject, but not religion.

Observing Discrimination

Bhagwati (1973) makes one of the earliest notable contributions to the empirical study of gender and development in his exploration of the relationships between gender, education, employment, and income equality within India. He notes systematic disparities in outcomes for females in India. While Bhagwati makes useful progress in considering gender and its relationship to outcomes of interest, gender, per se, is not the primary focus of his research, which has shifted more towards arguments about the relationship between trade and poverty (Bhagwati 2004). Though an early contributor to the empirical study of gender outcomes in India, Bhagwati did not develop gender as a focus of his work during his fruitful research career lasting multiple decades.

Hindered by the lack of intra-household expenditure data, Deaton (1989, 1997) analysed spending patterns at the household level to make inferences with regard to the relationship between spending patterns within the household and gender in India. Deaton used expenditure patterns associated with “adult” goods, including alcohol, tobacco, and adult clothing to measure whether households spent more or less on adult goods when the household had either a boy or girl. From this relationship, he was expecting to be able to observe intra-household bias from household-level data, though his results were not statistically significant. However, it should be noted that the absence of evidence does not constitute evidence of absence of an effect. Deaton (1997) emphasises that although research often concerns the individual, most of the data that exists is recorded at the household-level. Therefore, accurately obtaining outcomes of interest pertaining to discrimination against females such as intra-household spending patterns on health and education tends to be challenging. School attendance rates would be another individual-level outcome that would be revealing if it were accurately recorded on a large scale.

Deaton (2010) takes issue with some forms of experimentation, particularly randomised controlled trials (RCTs), for not adequately considering the underlying mechanisms driving outcomes, specifically with regard to economic development. My work is distinct from the projects Deaton rightfully critiques in that: (i) my project uses a survey experiment, rather than an RCT; (ii) I analyse potential mechanisms driving the outcomes I am able to measure, specifically with regard to discriminatory attitudes toward the movement of females; and (iii) my project does not focus directly on economic development or associated interventions (as opposed to mechanisms). Even though Deaton takes issue with certain uses of experimentation his work informs and inspires my own, which makes use of experimentation in a different application.

Sen has, needless-to-say, made a number of important contributions throughout his distinguished career to the empirical study of discrimination against women in India and beyond. His 1983 article with Sunil Sengupta in the Economic & Political Weekly reports evidence of systematically higher rates of malnutrition in girls relative to boys, although both boys and girls had high rates of undernourishment (Sen and Sengupta 1983). The analysis was nuan­ced, but the sample was limited to only two villages, allowing criticism associated with external validity. Like Deaton, Sen (1987) emphasises the need to move beyond the household as the level of analysis, to the individual, with a particular focus on the deprivations of women relative to men; Sen rightly argues gender as a critical parameter of analysis both outside, and, importantly, inside the household.

Sen (1990) famously proclaimed more than 100 million women to be “missing” throughout the world. The problem of missing women is particularly acute in India, where the population is close to 1.3 billion and the sex ratio, according to the 2011 government census, is 940 females per 1,000 males. In Development as Freedom, Sen (1999) considers various aspects of the relationship between gender and development, particularly with regard to the relationship between women’s agency and social development outcomes associated with health and education of children. Sen has helped to shine a light on the unacceptable disparities associated with gender as well as to document the positive externalities associated with enhancing the capabilities of women.

Drèze and Sen (1995, 2002) empirically examine various aspects of female deprivation within India. They examine sex ratios over time and by state, religion, and caste. They also present evidence for the many appealing downstream consequences of female education as well as negative externalities associated with education deprivation. Drèze and Sen suggest that women’s agency is a “force for change” that is consistently neglected in the development literature. While Drèze and Sen point out the externalities associated with women’s agency, they also recognise women’s agency as intrinsically desirable, claiming the depth and persistence of gender disparities to be one of India’s gravest social failures that will not necessarily decline with economic growth. Moreover, female agency is endogenous to cultural factors like patriarchy, which restrict economic development more broadly.

Murthi et al (1995) argue convincingly that aspects of women’s agency, particularly with regard to labour force participation and literacy, have significant effects on child survival. Moreover, Drèze and Murthi (2001), using district-level panel data, find evidence that women’s education is an important determinant of the variation in fertility rates within India. Drèze and Geeta Gandhi Kingdon (2001) explore the determinants of school participation within rural North India and find convincing evidence for the impact of household variables, particularly with regard to parental education, on school participation for children, both in general and for female children more specifically. My work builds on theirs by considering the relationship between education and discriminatory attitudes towards the mobility of females, which has further implications for school participation.

Esther Duflo has made useful contributions to the empirical study of gender and development in India, particularly with regard to the application of experimental methodology. Duflo, Abhijit Banerjee, and their team at the Abdul Latif Jameel Poverty Action Lab, have made important strides in incorporating randomised-controlled trials to the application of development research. However, there is considerable opportunity to further explore various aspects of gender and development using experimentation in innovative ways. In the first of its kind, my project uses a survey experiment to both empirically observe discrimination against women in the context of mobility at the individual level, and explore education as a possible mechanism associated with such discrimination.

Outside the specific context of India, few have done more than Rose McDermott to influence the field of political science, and the social sciences more broadly, with regard to the use of experimental methods. McDermott (2002) discusses the relative advantages and disadvantages of experimentation, surveys the use of experimentation in several social science disciplines, and advocates for the increased use of properly designed experiments in the field of political science. McDermott, in her work with John Gerring, outlines ways in which experimental methods might be more effectively integrated into case study research in the social sciences (Gerring and McDermott 2007). McDermott (2011) lays out a nuanced consideration of concerns associated with the external validity of experimentation in political science.

McDermott has also made vital contributions to the empirical study of women. Her work with Valerie Hudson et al (2009) on the WomanStats project has made great strides in providing access to data for the empirical study of the treatment of women across countries.1 Hudson, McDermott, and others convincingly argue that there is a relationship between the treatment of women within a state and its security. McDermott and Jonathan Cowden (2014) demonstrate the relationship between polygyny (that is, one man with multiple wives) and multiple undesirable outcomes, including increased rates of domestic violence, poor education outcomes, differences by gender in rates of HIV infection, as well as political rights and civil liberties.

Mobility of Women

The freedom of movement has been recognised as an essential right of women, as codified in Article 15(4) of the Convention on the Elimination of All Forms of Discrimination Against Women (CEDAW).2 However, physical restrictions on the freedom of women persist in many forms with important implications for the opportunities of women in a number of spheres, including health, education, recreation, political involvement, social engagement, and labour force participation. The freedom of movement is a vital aspect of female empowerment; understanding the extent to which discriminatory attitudes towards the freedom of mobility exist and the mechanisms underlying potential limitations on such freedom has useful implications.

This article builds on the existing work that empirically examines the treatment of women, particularly in India, by using an innovative survey experiment. Using a survey experiment allows one to empirically observe discrimination against females, compare outcomes between states, and examine potential determinants of variation, including education, gender, and religion. To do so, I will see how subjects respond to the question based on the treatment condition and, using a series of regressions, test for treatment effect heterogeneity—whereby the treatment condition of the question has a differential effect on response rates by subgroup—which may be disaggregated by level of education, gender, and religion of the respondent.

One of my central hypotheses is that the mobility of girls is systematically restricted, relative to boys, by their guardians. To test this hypothesis, I designed an experimental survey question whereby I invert the gender of the subject of the question while all other remaining details remain identical, so that I can empirically observe whether or not such discrimination exists. If, indeed, responses to the substantive question vary significantly based on treatment condition, it may be considered an indication of discrimination if respondents are statistically less likely to answer in the affirmative when the subject of the question is female. I discuss the construction of the question in further detail below.

I also expect to observe regional variation in the intensity of discrimination against females. To test this hypothesis, I will disaggregate my analysis of the treatment effect by state. I suspect the treatment of women to be worse in states like Bihar and Punjab relative to Kerala. Outcomes for women (for example, sex ratio, various health and education indicators, labour force participation, etc) tend to be much better in Kerala compared to Bihar or Punjab, which is likely to be reflected in the analysis of the treatment effect, again, indicating discrimination against women, by state. In other words, the treatment condition of the substantive question is more likely to influence responses in ­Bihar or Punjab relative to Kerala, though the treatment condition might still be significant in Kerala.

Another hypothesis is that education might be associated with variation in observed patterns of discrimination. For this hypothesis, I will test for treatment effect heterogeneity associated with level of education of the respondent using a logistic regression and response to the substantive question as the output. To test for treatment effect heterogeneity, it is first necessary to achieve a significant main effect associated with the treatment condition (that is, gender of the subject of the question), which, as mentioned above, might be considered an indication of discrimination against the mobility of females if subjects are more likely to respond in the affirmative when the subject of the question is male. The term “subject” can be used both in reference to the subject of the question with regard to sentence construction, as well as in reference to the survey respondent.

I suspect two other explanatory variables—which I will analyse for treatment effect heterogeneity—might also help to explain discrimination against women: gender and religion of the respondent. I suspect men to be more likely to discriminate against women. Religion may be associated with variation in discrimination against women, but my ability to test this empirically is limited with the data set such that it is. My analysis of
income is also restricted because my sample is heavily skewed towards the poor, as Bihar is the poorest state in India and much of my sample is drawn from rural populations.

Using a survey experiment in this context to empirically observe discrimination against women in any way in India has not been done before, to the extent that I am aware. This innovative approach allows the ability to empirically observe discriminatory attitudes and consider determinants by testing for treatment effect heterogeneity. The findings have implications for the study of variation in the treatment of women, and contributes to the growing literature empirically exploring discrimination against females.

Methodology

Between 2013 and 2016, I conducted the survey experiment in the states of Bihar, Punjab, and Kerala, as well as the city of Delhi, involving more than 5,300 subjects. The samples are drawn from four districts within each state and, likewise, the sample in Delhi was drawn from the four quadrants of the city. At least 1,200 subjects from each state were sampled with the goal of achieving a representative sample for each state by gender, religion, education, income, and so forth. A representative sample is a defining feature of a survey experiment and allows the ability to make inferences with regard to the association between the effect of the treatment condition and subgroup at the state-level as well as the ability to compare results between states.

A survey experiment allows the possibility to empirically observe biases in attitudes, in this case having specifically to do with the mobility of females. A survey experiment offers an innovative way to both empirically observe discriminatory attitudes towards the freedom of movement of females, as well as the opportunity to analyse potential determinants of variation in such attitudes, stemming from discrimination. A significant main effect will indicate discriminatory attitudes toward the freedom of movement of females. The main effect can then be tested for treatment effect heterogeneity whereby the treatment effect has a differential effect on subgroups, in this situation, potentially having to do with gender, education, or religion of the respondent. My hypothesis is that all three explanatory variables might potentially be associated with variation in discriminatory attitudes towards females.

The experimental treatment associated with the question analysed in this article has to do with the gender of the subject of the question; in one version of the question, the subject is a male, while the subject in the other version is female, while all other details of the question remain identical. Respondents are unaware that an experimental treatment exists or that there is another version of the question. Beyond the treatment condition, I would like to highlight that the safety of the subject of the question is specified in the vignette in an attempt to mitigate potential response bias associated with differential safety concerns associated with gender. The two versions of the question are presented below, with the experimental manipulation highlighted by quotation marks.

(A) Kumar’s 16-year-old “son” was awarded a voucher/certificate to travel to a historical site and participate in a lecture of a famous historian. In order to travel to this historic site, Kumar’s “son” has to travel alone for about an hour by public transportation. The route to the site is safe and there is regular transportation available. Do you think that Kumar should allow his “son” to travel alone?

a Yes

b No

(B) Kumar’s 16-year-old “daughter” was awarded a voucher/certificate to travel to a historical site and participate in a lecture of a famous historian. In order to travel to this historic site, Kumar’s “daughter” has to travel alone for about an hour by public transportation. The route to the site is safe and there is regular transportation available. Do you think that Kumar should allow his “daughter” to travel alone?

a Yes

b No

The main effect is the term used for the significance of the treatment condition, or experimental manipulation. In a survey experiment, if the main effect is significant, the treatment condition (that is, explanatory variable) has a statistically significant effect on average response (that is, dependent variable) to a particular substantive question; the treatment condition “causes” a difference in average response rates. For this question, the treatment condition has to do with the gender of the subject of the question in the context of a situation that has to do with attitudes towards mobility.

A challenging aspect to the design of this question is the avoidance of a potential bias in responses associated with safety concerns. However, it is unlikely that education might be associated with “decreased” concern for the safety of girls; if education is associated with an increased likelihood of answering in the affirmative, it is unlikely that safety would be driving the main effect. In other words, if a heterogeneous treatment effect is observed with regard to education, it would be improbable that the main effect is being driven by safety concerns. An admitted flaw in the experimental design of the question is that, in the question, the subject is going to an education event; in hindsight, I would probably have made it another type of event so as to not have any possible contamination effect associated with bias against the education of females, which might detract from the effect of mobility restrictions, which is what I am trying to observe.

Results

I will now consider the main effect, or treatment effect, of the experimental manipulation, which is an indication of the average difference in response rates associated with the experimental manipulation. The main effect is the average causal effect of the treatment condition on response rates. Table 1 is the OLS regression output showing the main effect associated with the significance of the treatment condition, indicating that subjects are, on average, 15% less likely to respond to the question in the affirmative when the gender of the subject of the question is female.

Figure 1 shows the significance of the main effect, or treatment effect. ­Respondents are significantly and substantially more likely to answer in the affirmative if the subject of the question is male. The main effect (that is, how subjects answer the question differently as a function of the treatment condition) indicates strong discriminatory attitudes towards the freedom of mobility of females. Subjects (respondents) are approximately 15% less likely to respond to the question in the affirmative when the gender of the subject of the question is female.

The output associated with the regression above indicates discriminatory attitudes of respondents towards the public mobility of females. The dependent variable is response (that is, “yes” or “no”) to the question of whether the respondent thinks the subject should be able to attend the desirable event described in the question, while the independent (that is, explanatory) variable is the treatment condition. Of critical importance is whether or not the treatment condition significantly influences average response rate, which indicates discriminatory attitudes towards the mobility of females. A relative difference in response rates as a function of experimental treatment is considerably more important than absolute response rates or probability that, overall (regardless of the experimental treatment), respondents answer in the affirmative. From the regression, it is clear that the experimental condition is unambiguously significant. The regression coefficients in Table 2 represent the difference in the estimated probability that a respondent will answer the substantive question one way or the other as a result of the experimental condition (for example, respondents in Bihar are 20% more likely to answer the question in the affirmative when the subject
of the question is male), disaggregated by state.

The relative magnitude of the coefficient associated with the main effect of the mobility question follows the pattern I would expect: strongest in Bihar, Punjab, and Delhi, weaker—though still significant at the 90% confidence interval—in Kerala. According to a number of indicators, Bihar and Punjab are among the worst regions in India to live for females, which is corroborated in my findings. Other likely regions where the freedom of women remains most restricted in India would likely include Uttar Pradesh (the population of which is more than 100 million, which would make it about the size of Brazil and one of the 10 largest countries by population in the world), Haryana, Rajasthan, Jammu and Kashmir, and several other provinces, particularly in northern India.3

I will now turn to the relationship between education and discriminatory attitudes towards the mobility of females. First, I will use an ANOVA model to consider the significance of the interaction between the treatment condition and education of the respondent. Table 3 shows the ANOVA model of the independent variable, edug, showing a significant statistical interaction between education and predicted response rate for the data from Bihar, Punjab, Kerala, and Delhi.

To better understand the relationship between education and discrimination, I will analyse treatment effect heterogeneity with regard to level of education (that is, less than primary, less than secondary, and at least secondary). I will see if the experimental manipulation affects responses differently based on level of education. Using a logistic regression, I predict the probability of response type (that is, affirmative vs negative) for education level relative to experimental treatment in an attempt to determine if education might mitigate discriminatory attitudes toward the mobility of females. Figure 2 shows treatment effect hetero­geneity associated with the level of education of the respondent for the entire data set (that is, samples from Bihar, Punjab, Kerala, and Delhi). Education is associated with a significant reduction in discriminatory attitudes towards females in the context of mobility.

Figure 2 depicts treatment effect heterogeneity with regard to level of education, which indicates a relationship between education and the effect of the experimental treatment on average response rates. In the analysis, level of education is appropriately coded as a dummy variable for each level of education, so the effect of each level is considered independently. The findings indicate that education is associated with variation in predicted response rates for the female treatment condition, but not the male treatment condition. In other words, predicted response remains statistically constant across levels of education for the male treatment condition, but not for the female treatment condition, indicating that level of education is significantly associated with predicted response for the female treatment condition. Thus, education is significantly associated with discriminatory attitudes towards females.

Gender of the respondent is also associated with average variation in discriminatory attitudes towards females, even though both genders display discriminatory attitudes. Still, the average effect of the treatment condition on response is greater for men than women: men display higher rates of discrimination against females in the context of this substantive question having to do with mobility. The Anova model considering the interaction of the treatment condition and gender of respondent is included in the Appendix (p 62). Figure 3 shows the difference in average response rates based on experimental condition and disaggregated by gender; the magnitude of the main effect is larger for male respondents (right column), though the treatment effect is still significant for female respondents (left column).

 

 

Religion is another obvious possible explanatory variable driving variation in discriminatory attitudes and behaviour towards women. Might religion account for the patterns of discrimination against women within India? Since my sample draws heavily from Punjab, it is possible to compare the treatment effect for Sikhs relative to Hindus or Muslims, though my analysis is limited by the size of subgroups within the sample. In the context of this question, religion is associated with an intercept shift of the treatment effect rather than a significant difference in slope of the regression, indicating that the observed bias associated with the experimental effect is not significantly different for Sikhs relative to Hindus or Muslims, though each subgroup displays discriminatory attitudes with regard to the experimental treatment in the question. While Sikhs, Hindus, and Muslims display discriminatory attitudes with regard to the mobility of females, a larger sample is needed to be able to confidently make inferences with regard to religion as an explanatory variable of variation in discriminatory attitudes towards females.

Discussion

I began by asking whether it is possible to empirically observe discriminatory attitudes towards the mobility of females within India. I found, using a survey experiment, that indeed the main effect was significant, though the magnitude of the average effect on response rates varies by state, particularly with regard to Bihar and Punjab relative to Kerala (though the main effect remains significant in Kerala as well), with Delhi falling somewhere in between. I then asked whether a heterogeneous treatment effect was associated with education, gender, or religion of the respondent. I found treatment effect heterogeneity associated with level of education and gender of respondent, but the results are inconclusive for relative comparisons between religions.

The main effect for the experimental treatment in the substantive question considered in this article unambiguously influences average response; the experimental manipulation associated with the gender of the subject of the question causes a difference in average response rates. Moreover, there is significant treatment effect heterogeneity associated with level of education; education is associated with a reduction in the magnitude of the treatment effect. Gender of respondent is also associated with a heterogeneous treatment effect; the treatment condition is more likely to affect response rates for men relative to women.

Restricting the public mobility of females has immediate implications on well-being, in addition to reflecting broader patterns of the subordination of women. Gender-based limitations on mobility systematically restrict the access of females to a wide spectrum of activities and experiences associated with quality of life, including health, education, political involvement, social engagement, and labour force participation. Restricting the mobility of girls is a form of control that has both immediate and longer-term implications. Over time, limitations on the mobility of girls affects their education and, in turn, their ability to have greater control over resources, which is likely to have further implications for susceptibility to male dominance more broadly.

My work indicates that the potential survey experiments have to explore aspects of discriminatory attitudes and their determinants in the context of not only gender, but also with regard to caste, class, and religion. Such exercises might help to empirically observe and better understand variation in outcomes both within India as well as in the international comparative context. Reducing disparities in outcomes along economic, social, and physiological lines will remain a critical global challenge for the foreseeable future.

Notes

1 WomanStats Project Database (2017), http://www.womanstats.org.

2 http://www.un.org/womenwatch/daw/cedaw/text/econvention.htm#article15.

3 I would like to make clear that I believe the distinction between the treatment of women in North India relative to South India to largely be an oversimplification (that is, north vs south). Some of the worst outcomes for social and economic indicators in the world are found in regions of northern India, though discrimination against women persists in southern India as well.

References

Anand, Sudhir and Amartya Sen (1995): “Gender Inequality in Human Development: Theories and Measurement,” Human Development Report Office, United Nations Development Programme, New York.

Bhagwati, Jagdish (1973): “Education, Class Structure and Income Equality,” World Development, Vol 1, No 5, pp 21–36.

— (2004): In Defense of Globalization: With a New Afterword, USA: Oxford University Press.

Deaton, Angus (1989): “Looking for Boy–Girl Discrimination in Household Expenditure Data,” The World Bank Economic Review, Vol 3, No 1, pp 1–15.

— (1997): The Analysis of Household Surveys: A Microeconometric Approach to Development Policy, World Bank, Washington DC.

— (2010): “Instruments, Randomization, and Learning about Development,” Journal of Economic Literature, Vol 48, No 2, pp 424–55.

Drèze, Jean, and Geeta Gandhi Kingdon (2001): “School Participation in Rural India,” Review of Development Economics, Vol 5, No 1, pp 1–24.

Drèze, Jean and Mamta Murthi (2001): “Fertility, Education, and Development: Evidence from India,” Population and Development Review, Vol 27, No 1, pp 33–63.

Dreze, Jean and Amartya Sen (1995): India: Economic Development and Social Opportunity, Oxford: Oxford University Press.

— (2002): India: Development and Participation, Oxford: Oxford University Press.

Gerring, John and Rose McDermott (2007): “An Experimental Template for Case Study Research,” American Journal of Political Science, Vol 51, No 3, pp 688–701.

Hudson, Valerie M et al (2009): “The Heart of the Matter: The Security of Women and the Security of States,” International Security, Vol 33, No 3, pp 7–45.

McDermott, Rose (2002): “Experimental Methodology in Political Science,” Political Analysis, Vol 10, No 4, pp 325–42.

— (2011): “Internal and External Validity,”Cambridge Handbook of Experimental Political Science, pp 27–40.

McDermott, Rose and Jonathan Cowden (2014): “Polygyny and Violence against Women,” Emory Law Journal, Vol 64, No 6, pp 1767–814.

Murthi, Mamta, Anne-Catherine Guio and Jean Drèze (1995): “Mortality, Fertility, and Gender Bias in India: A District-level Analysis,” Population and Development Review, pp 745–82.

Sen, Amartya and Sunil Sengupta (1983): “Malnutrition of Rural Children and the Sex Bias,” Economic & Political Weekly, pp 855–64.

Sen, Amartya (1987): Gender and Cooperative Conflicts, No 1342, Helsinki: Wider.

— (1990): “More than 100 Million Women Are Missing,” New York Review of Books, Vol 37, No 20, pp 61–66.

— (1999): Development as Freedom, New York: Alfred A Knopf.

Subramanian, Shankar and Angus Deaton (1991): “Gender Effects in Indian Consumption Patterns,” Sarvekshana, Vol 14, No 4, pp 1–12.

Updated On : 7th Mar, 2019

Comments

(-) Hide

EPW looks forward to your comments. Please note that comments are moderated as per our comments policy. They may take some time to appear. A comment, if suitable, may be selected for publication in the Letters pages of EPW.

Back to Top