Is it Better to be a Boy? A Disaggregated Outlay Equivalent Analysis of Gender Bias in Papua New Guinea

Abstract

The demographic evidence of gender bias in many countries has provided an impetus for finding ways to study the status of women in developing countries. Because of the lack of accurate intra-household data, Deaton [1989] introduced a method for using household expenditure data to infer discrimination in the allocation of goods between boys and girls. Few studies of discrimination using the method, however, have detected bias even though alternative indicators suggest it is a serious problem. In this paper, we study the case of Papua New Guinea, a country in which there are many indicators of severe gender bias. Discrimination in the allocation of goods between boys and girls within households in Papua New Guinea is examined using Deaton's outlay-equivalent ratio method. Adding a boy to the household reduces expenditure on adult goods by as much as would a nine-tenths reduction in total outlay per member, but girls have no effect on adult goods expenditure. The hypothesis of Haddad and Reardon [1993] that gender bias is inversely related to the importance of female labour in agricultural production is not supported. Sensitivity analysis shows that bias in rural areas occurs equally regardless of the age of the household head, while bias against girls may be less in regions of the country that have ethnic groups which practice matrilineal descent.

Notes

John Gibson, University of Waikato. Address for correspondence: Department of Economics, University of Waikato, Private Bag 3105, Hamilton, New Zealand. E-mail: [email protected]. Scott Rozelle, University of California, Davis

There appears to be some ambiguity in the statistics, because the 1996 Demographic and Health Survey reports an under-five mortality rate of 108.2 for males and 91.3 for females, which is the opposite to the pattern reported in the WHO statistics [ NSO, 1997 ].

In fact, according to the UNDP's gender-related human development index, the position of women in PNG while not good, is better than a number of other countries, such as Burkina Faso, Bangladesh and Côte d'Ivoire. While this seems to contradict the arguments of the social scientists that claim there is a lot of discrimination, it may be because the index weights some of the opportunities that PNG women face in a way that makes the measure look lower relative to other countries. It could also be that the index is not a very good measure of human development. Case and Deaton [ 2002 ], for example, note that the UN's gender-related index misses a number of the deprivations that are associated with poverty.

In other words, adult goods consumption in the present may be affected both by current income and by expected lifetime income. If having girls in the family leads to an increase in expected lifetime income due to, say, bride price receipts, it may lead to a positive impact on adult goods consumption which is independent of any impacts due to discrimination. The outlay-equivalent literature does not consider the fact that the demographic structure of the household may affect expected lifetime income, with the result that the demographic variables may pick up omitted effects in addition to any discrimination that is present.

Unobserved household-level heterogeneity that affects consumption patterns should not affect the inferred level of bias, unless such heterogeneity affects the demographic coefficients for boys differently than for girls.

This classification was established from an agricultural mapping project [ Allen, Bourke and Hide, 1995 ].

The monetary values for self-produced foods were the values used by respondents. Estimates of average expenditure are unchanged if these respondent-reported unit values are replaced by either cluster medians of the unit values or cluster averages of market prices [ Gibson and Rozelle, 1998 ].

We are grateful to Roberto Gutierrez of Stata Corporation for suggesting this procedure. Because repeating the same observations induces a correlation in the disturbances of the stacked model, the clusters also have to be redefined as the interaction of the original clusters with the dummy variable for each replication of the dataset.

Specifically, we used an adjusted Wald (W) test for zero slopes: (d−k + 1/kd)W∼F(k, d−k + 1), where d is the number of clusters minus the number of strata (105), and k the number of slope variables [ StataCorp, 1999 ].

Three of 32 π-ratios were positive in the results for Côte d'Ivoire and 8 of 48 in the results for Thailand, which included some items such as ‘men's and boy's clothing’ which were known to include child goods.

Actually, there may be a second, although in our opinion, more minor concern. In contrast to the results in Deaton [ 1989 ], the equality of the π-ratios across goods for each of the adult demographic groups is also not rejected. While this equality is less informative for adults than it is for children, the failure to reject may indicate low test power, although once again, if the survey design effects are ignored the equality across the goods is rejected at the 0.02 level, for at least two of the adult groups.

If the survey design effects are ignored, as they seem to have been in some previous studies, the p-value for 7–14 year old girls would also fall into the rejection zone, at p = 0.02 (from 0.12 with survey design effects included).

The importance of controlling for survey design effects can be seen by noting that if the test of equal π-ratios for the effect of older boys and older girls on betelnut (alcohol) consumption is carried out with the clustering and stratification ignored, the p-values fall from 0.08 to 0.02 (0.30 to 0.13).

The p-value for the difference between π-ratios for prime age males and females in the aggregate adult goods group drops from 0.28 to 0.06 if betelnut is excluded, probably because adult women are equal participants in consumption of betelnut in contrast to alcohol and tobacco which are predominantly consumed by men.

In fact, child labour for cash incomes is quite rare in Papua New Guinea. Children between the ages of 11 and 18 may work in family-related enterprises if they have parental permission, a medical clearance and a work permit from a labour office. But according to the 2001 Report of Human Rights Practices, from the US Department of State, such employment is uncommon in PNG.

This equality across the two rural zones appears to be robust to the definitions of high and low income areas. If the lowest of the clusters in the high agricultural income potential zone are reallocated to the low income zone, and the outlay equivalent ratios re-estimated, the hypothesis of equality of π-ratios across the two rural zones is still not rejected (p < 0.45).

These include the Trobriands, the Tolai, the Kavieng, and various groups in Milne Bay and Bougainville. Within the household survey, these areas are identified as coming from the Esa'ala, Kavieng, Kokopo, Losuia and Rabaul districts.

However, Quisumbing and Otsuka [ 2001 ] report that in matrilineal areas of Sumatra, daughters are still disadvantaged in terms of schooling.

Additional information

Notes on contributors

John Gibson

John Gibson, University of Waikato. Address for correspondence: Department of Economics, University of Waikato, Private Bag 3105, Hamilton, New Zealand. E-mail: [email protected]. Scott Rozelle, University of California, Davis