Presentation on theme: "1 Why do we think that education of women the nearest we have to a magic bullet in development? Richard Palmer-Jones School of International Development,"— Presentation transcript:
1 Why do we think that education of women the nearest we have to a magic bullet in development? Richard Palmer-Jones School of International Development, University of East Anglia paper presented at DESTIN, LSE, 26/2/10
2 Educating females is the single most effective development intervention – an (almost) undisputed orthodoxy of the past 25 years: –Two prominent examples Duncan Green, Head of Research, Oxfam Amartya Sen - female autonomy & empowerment – Development as Freedom; The Idea of Justice –Womens agency as intrinsic and instrumental to well-beings of all – particularly: »fertility, infant survival, child quality (nutritional status and educational attainments), etc. –Important, convincing and well institutionalised development agenda But leads to misdirected policy and projects –Despite more than 25 years of advocacy, and intervention, the juvenile sex ratio in India has become more adverse to females – more missing women! So do we believe these arguments more than reason warrants? –leads to accepting relatively uncritically evidence of causal links between female education and beneficent development outcomes Wastes opportunity to try to do better and implicitly kills and maims persons Illustrate with 4 examples –Whatever happened to fertility in Nigeria?- John Caldwell, 1979 –Does gender inequality in education harm growth? - Stephen Klasen (,2002, & 2009) –Are females are better transmitters and receivers of literacy externalities? - Basu & Foster, 1998 & Basu, Narayan & Ravallion, 2001,.. –Does educating females reduce fertility, infant mortality, & gender discrimination in India? Murthi, Guido & Dreze, 1996, and Dreze and Murthi, –These ill founded ideas lead to the production and academic plainly publication of silly things (and reproduction in the media) Have heights of adult males have increased faster than those of adult females in recent cohorts in India? - Deaton, 2008 Lessons of this sorry tale The argument:
3 Duncan Green, Head of Research, OXFAM
4 Questioning the orthodoxy –Female education and fertility reduction in Nigeria Caldwell, 1979 –Rising female education has not greatly improved child well-being –And partners education (and ethnicity – not religion) also counts »assortative mating, labour market mechanisms, and unobservables –Gender and Growth Female education is good for growth (Klasen, 2001;Klasen & Lammana, 2009) –Not when you control for institutions and the resource curse (Papyrakis & Palmer- Jones, 2009) –Proximate literacy in Bangladesh (and India) Basu & Foster, 1998; Basu, Narayan and Ravallion, females are better recipients as well as better transmitters of literacy externalities –Selective reporting of results, which do not survive in other data sets, and fail to exclude alternative hypotheses (Iversen and Palmer-Jones, 2008) »Its the type of man who chooses to marry educated women, stupid! –Sex Ratios in India Fertility, mortality and female disadvantage reducedby females education (and labour force participation) –Murthi, Guido and Dreze, 1996 & Dreze & Murthi, 2001 –Culture rather than female autonomy drives beneficent outcomes –Heights of adult males increased faster than females heights in India Deaton, 2008 – adult male heights have increased faster than female –Who would use changes in absolute heights? Duh…. –Why question this orthodoxy? slings
5 Since 1970s this orthodoxy has been almost uncontested –Caldwell in Nigeria (1979) –[T]he preceding analysis has shown that maternal education is the single most significant determinant of these marked differences in child mortality (Caldwell, 1979:408) –But UN,1985, Socio-Economic Differentials in Child Mortality The other variable with effects comparable in magnitude to mothers education is fathers education, …. ( Table II.13 & Table III.6 ) Average effect of an additional year of mothers and fathers schooling on child mortality (stage II*) Living in: Schooling of: RuralUrban Mother Father * Seems to be an error – Table II.13 & Table III.6 show these figures to be Stage III results; stage III is stage iii run for urban and rural samples separately Stage II regressions include all other variables (p10)
6 Many more cases of more educated partner than carer What are the likely implications of there being a higher proportion of educated males than females at each level of education, and what are the implications for the regression coefficients? What if there is a non linear (declining) relationship between outcomes and levels of education (as measured)? (spine plot)
7 ( for DID estimation see Osili & Long, 2008 ) Carers & partners education equally significant partners education changes sign when wealth score included Results not significant Or only on years of education not fertility
8 Partners education is just as important as carers Fertility has not decreased greatly in South West (area of Caldwells research) despite increases in female (and male) education
9 Selective estimation and reporting; Difficulties of replication with same data sets, and getting same results with other similar datasets Neglecting assortative mating and reverse talents effect – men are more educated than their partners but tend to be less able at each level of education;
10 Education externalities Females are better transmitters –Effective Literacy (Basu and Foster, 1998) Literacy rate = L/N (L = literates, N = Population) Effective literacy = (L+e(N-L))/N, 0 e m where superscripts reflect gender of literates (e.g. e f = being female-proximate, or proximate to a female literate; e m similarly) and recipients of literacy externalities! –Proximate-illiteracy externalities (Basu, Narayan and Ravallion, 2002) Wages of female proximate-illiterates in non-farm employment greater than those of isolated female illiterates (Bangladesh HIES, 1995/6) e f > e m where subscripts are gender of illiterates (e f = female proximate-illiterate, e m similarly) Basu and Foster, 1998, Economic Journal; Basu, Narayan, and Ravallion, 2002, Labour Economics
11 Female proximate-illiterates are generally male-proximate –Hence if e f > e m then e m >= e f E.g. if females are better recipients then males are at least reasonable transmitters Selection coefficient on household literacy is negative –I.e. the female labour force participation of illiterate females is lower in literate households Presumably an adverse to (illiterate) females outcome since they are denied empowerment in the wage labour market? Female illiterates are more likely to be proximate to a male than a female literate
12 Female proximate-illiterates in off-farm employment have negative externalities of child nutritional status Non-replicability and misleading reporting of proximate- illiteracy findings (B/d HIES 1999/2000; Indian NSS CES, various years) –Replication in econometric studies? Prominent papers which are framed as supporting female literacy externalities have simple flaws: –Gibson, 2001, and Alderman et al., 2003, show greater positive child nutritional externalities of female-proximate illiteracy but use community proportions of female literates with limited controls on community characteristics details in Iversen and Palmer-Jones, 2008, JDS, 44(6):
13 Gender inequalities in education and economic growth –Gender inequalities in education are bad for growth Klasen, 2002, World Bank Economic Review, and Klasen and Lammana, 2009, Feminist Economics –Based on cross-country regressions Are gender-growth relations different in resource curse economies? –Resource curse economies have particularly low growth & –low female empowerment indicators Gender-growth orthodoxy neglects institutions and political geography –Institutional variables eliminates significance of increase in relative female education –Maybe only when institutions are right is female empowerment effective?
16 Low female education Low female labour force participation Low female entrepreneurial earnings Low economic growth Poor female human capital Poor child human capital accumulation Low female political participation Klasen type explanatory framework
17 Weak institutions Dutch disease Resource curse Relative decline of agriculture & tradables Rise in male earnings Rise in government rents & transfers Fall in female wage/entrepreneurial earnings Rise in female opportunity cost (unearned income) Fall in female labour force participation Fall in female political participation/ representation Grabbing politics Appropriable resources Adapted to resource curse
18 Adapted to resource curse and institutional specificities
19 Macro-economic results
20 Dreze, Sen & Womens agency –Employment –Education –Bargaining power –Perception and status –Entitlements –In his analyses of Indian data Sen relies heavily on Muthi, Guido & Dreze, 1996, and Dreze & Murthi, 2001 – e.g. Sen, 1999, p194 fn 14, 16, 17 – also refers to John Caldwell, it is also the case that the limited role of womens active agency seriously afflicts the lives of all people focusing on womens agency … precisely [because of] … the role that such agency can play in removing iniquities that depress the well-being of women …variables such as womens ability to earn a independent income and find employment outside the home, … to have literacy and be educated participants in decisions … has more voice … employment often has useful educational effects.. The diverse variables identified in the literature thus have a unified empowering role.. (Sen, Development as Freedom, 1999:191-2; emphasis in original)
21 Sen. 1999: There is considerable evidence that womens education and literacy tend to reduce the mortality rates of children (195)… p196 relies on Murthi et al oevre)
22 Fertility, Child Mortality and (Juvenile) Sex Ratios in India –Adverse to female sex ratios historically –Female neglect/infanticide/foeticide North-south differences (decreasing) –wheat/rice – economic value of female labour (Bardhan) –Culture & kinship (Dyson & Moore) »Hypergamous & exogamous Indo-Aryan north vs endogamous endogamous Dravidian south –Sopher, Miller, Kishore, Murthi,, et al., Agnihotri, Croll, …and many others –Basic argument is that less adverse outcomes are driven by female autonomy, by which they mean education (literacy) and waged employment
23 The problem is that despite increases in female employment and female literacy, the juvenile sex ratio has deteriorated between 1991 & 2001 Censuses Disregarding fertility decline (intensification effect) & spread of sex-selective abortion/foeticide
24 Source: Guilemotto, 2007 Female to male ratio 1961, 1981, 2001
25 However …. –Need to use juvenile sex ratios, especially 5-9 (excess male migration among youth and adults) Excess male infant deaths 0-1 means that 0-4 age range confounds excess male infant deaths 0-1 (0-3 months) with excess female deaths 1-4 (actually 4-48 months) Ethnic differences especially STs & SCs –STs - no/little discrimination apparent against girls but »higher infant deaths -> pro-female sex ratios –SCs – bias against females but high infant mortalities results in »apparenlty lower anti-female juvenile sex ratios –E.g. confounding poverty and gender bias effects –Muslims not showing gender bias Census has 20-30% missing infants 0-1, 1-2 compared to 2-3 ….. Due to (sex biases?) under-reporting Is female literacy more effective than male? Are literacy and employment confounded with culture (proxied by language) –Basic issue of causality or correlation Agnihotri, PJ & Parikh, 2002….
27 Sen. 1999: There is considerable evidence that womens education and literacy tend to reduce the mortality rates of children (195)… p196
28 VariableVIF1/VIF tflit15ppc tmlit15pc flpmw15ppc south east scpc stpc west hcr med_pc urbpc Mean VIF2.69 replication collinearity
29 Problems –Numbers of districts in MGD are 296, while we have 335 (or 332 in spatial estimations due to districts with two locations) 1981 census has 366 Districts Only use districts in main states –Excluding Jammu & Kashmir, Himachal Pradesh, North East States & Assam (no census in 1981), and Union territories –326 districts.Why exclude HP (or J&K) in 1981? »Seemingly HP has missing poverty line data as not in figure 1, but this does not apply to J&K which is? –A further 30 districts are lost »which ones are not reported »Bias?
31 Little difference between effects and significance of female and male literacy
33 First rule of rhetoric – ignore alternative explanations (Sheila Ryan Johannson, ….) –Is there any variable representing culture? Spatial dummies –MGD use regional dummies – what do these represent? »South, and West vs North –Agnihotri invents the kinship variable representing Indo-Aryan culture »Disadvantageous hypergamous, exogamous marriage »Based on informal classification of Districts by predominant language »Use predominant language data from 1961, 1971 and 1991 censuses to explain the kinship variable »The trick is to separate out the Hill Hindus (Berreman, …..) –How does one understand the very similar effects of female and male literacy? Assortative mating means households with educated females also have educated males –Who decides whether there is an educated female in the household? Husbands? –Alaka Basu, 1999, (in Bledsoe, ed., ) suggests it is the husband, or rather the husbands natal family which decides whether to marry an educated female into the family, and it is this rather than female education per se which drives the association of female education with beneficial outcomes.
The techniques of rhetoric Ignore dissenting views, positions, perspectives –Ignore problems raised by other scholars Refuse constructive dialogue or manipulate the context of debate –BURYING THE (BAD) NEWS – RHETORIC OR REALITY? Politician activism –But Who speaks for the muted? Does anything go, or is there (an only provisionally, fallibly known) reality? Construct convincing abstract model piling (dubious) assumption of (dubious) assumption Employ emotional hooks (frames) –Family, bereavement, injustice, explotiation, tragedy, disaster Use data selectively –Pick and choose – use carefully selected and misleadingly presented data –Build on extreme or unusual events/actions Homogenise phenomena –Treat unlike things the same – ceteris is not paribus Use inappropriate (but telling) metaphores or allegories
35 How seriously misled can you be by uncritical acceptance of gender bias? Figure 1 shows that … Indian men are.. [getting taller] at more than three times the rate of Indian women (Deaton, 2008, American Economic Review, 98(2):471
36 In India the starkest divisions are sometimes within the household. Indian women tend to have less clout than their African counterparts. Their claim on a family's resources may be weak, even as the demands made on them are heavy. Many women are consequently underfed or overworked during pregnancy. Their offspring, especially their daughters, are also undernourished during infancy. India may be growing taller as it grows richer. But, Mr Deaton shows, the average height of Indian men is rising three times faster than that of Indian women. NEW! Deaton's new paper on health, happiness and wealth is highlighted in The Economist
37 Really? With sample weights Without sample weights Absolute heights of males rise about a third faster than females Very odd
38 … some more problems, though Male sample is biased –NFHS report is not explicit how the male sample was chosen –DHS wealth index indicates bias Young males have higher wealth index than younger females and vice versa for older males and females opportunistic or convenience sampling? –Main sample is ever-married women »Unemployed male graduates and older labour dependent males? Including wealth index mitigates but does not eliminate female absolute height change disadvantage
39 Wealth and height of males and females Older men have lower z-scores Older women have higher z-scores Younger men have higher z-scores Younger women have higher z- scores
40 But, really, raw height is not the appropriate metric! –You should use z-scores of height! How to calculate adult height z-scores? –Needs an nutritionally/health unconstrained population as a standard & appropriate methods – you dont just pool the data … You are now entering the weird world of height measurement …. Cross section & longitudinal studies with height –Initially I used USAs NHANES3 Large sample, whites only …. many problems – white heights declining? (Komlos, et al.) –More recently I have been using England Health Surveys and other NHANES data from USA Both USA and UK height data vary between surveys & cohorts –USA heights of recent cohorts are unstable across survey years –In the UK heights of males increasing faster than those of females –Use UK longitudinal surveys NSHD – 1946 cohort; NCDS – 1958 cohort Unfortunately, some problems –Restricted access – eventually (only last week) received data –Often use self-reported heights –Measured heights at only a few ages –Some errors in measured data ….
41 Computing Z-scores from large cross- section surveys –Heights not normally distributed NCHS used different standard deviations above and below the median height at each age Distributions with variations in Location (Mean/median) and Shape (skewness/kurtosis) with age, sex and ethnicity (?) –New WHO standards use LMS/GAMLSS methods »LMSChartmaker (Excel macros) »GAMLSS suite – runs in R –To cut a long story short Use standard population (NHANES3) to estimate standard distributions by age (needs more recent data) Compute z-scores of observed heights from these estimated distributions.
42 Nhanes3 data - bumpy Nevertheless, ploughing on to see what it might be possible to say ….
43 Standardising Indian heights z-score (NFHS3) using distributions from NHANES 3 Column 1: raw height with age estimated by OLS; Column 2: z-score computed with NHANES3 estimated mean and sd for age Columns 3 & 4: z-score computed with LMS or GAMLSS estimates of height by age distributions Female height does not increase as much as male Female height increase as fast as male Female height increases faster than male
44 But, NHANES3 is unreliable Pooled whites born in USA only data from NHANES (to ) – with covariates education, PIR, region ….
45 Or use England Health Surveys ( ) Male Female Females reach maximum height earlier; males increasing in height over cohorts; problem of male 1940 born cohort
46 UK & USA data compared to Dutch (tallest population in the world) self-reported heights
47 Longitudinal studies by –Cline et al., 1989 –Sorokin et al used by Niewenweg et al., 2003; Webb et al., 2007 –Estimate height shrinkage/age functions –Varies with sex »Females start shrinking earlier and more Many (Deaton) who use adult height data assume no shrinkage till 50s –but this is cavalier Height probably starts shrinking around mid 30s, and is faster in women than men –(maybe earlier in poor populations?)
48 Depends on assumptions of maximum potential height – Waaler, 1984 suggests 185 cms for men and ?170? for women
49 Male and Female Mean Heights of Adults by Sex and Age (NHANES 3)
50 Compute z-scores from stylised height distributions –apply shrinkage equations with standard height distribution parameters Sorkin 1999a (because the cross country study shows heights increasing to mid 30s) Max male height at 23 – 183cm Max female height at 23 – 169 cms Coefficient of variation of height - 3.8%
51 Use shrinkage equations applied to guestimated potential maximum height Note: coefficient on female dummy sensitive to maximum height (location) but coefficients on Female * age are now negative and not sensitive to location Age shrinkage Females growing faster
52 Or use longitudinal studies – NSHD (7/3/1946 cohort) & NCDS (1958 cohort) NCDS measured data Height differences age 45-33
53 NSHD data (1946 cohort, measured data) Distribution of height differences, age
54 Differences in Self-reported heights Height lumping and inaccuracies
55 Problems in measuring heights –E.g. self reported height differences (ages 26 & 22) –According to Deaton difference in height rise between Indian males (0.5) and females (0.2) was about 0.3 cm per decade –Shrinkage estimate is about cm per decade (40s- & 50s) »Excess female shrinkage in.03 – 0.07 cms/decade Difference in measured heights between morning and evening average 0.6 cm –Eg diurnal variation can be a significant proportion of shrinkage –Untrained measurers of adult females (USA) (Lipman et al., 2006) Up to 18 cm difference from research assistant Even experienced measurers have significant variation –Standard deviation of height measurement about 0.09cm »95% obs within +/- 0.18cm –1-5 years experience cms mean difference from RA –>5 years experience - 2.4cms difference form RA –E.g. even for trained measurers (compared to a research assistant) measurement error can be a significant proportion of height change difference
56 These points are important because it reflects unreasonable acceptance of the female nutrition/health deprivation hypothesis –Reported in Sen of course – foetal undernourishment and low birth weight Cardiovascular diseases among south Asians (Osmani and Sen, 2002) Foetal undernourishment traced to female deprivation (LSE, 2010) –But is this deprivation relative to males, or to what they, females, could, ideally, be? –Partha Dasgupta (an inquiry into. & J of Econometrics, 1997) too poor not to discriminate (female foeticide among the Inuit) –But why does income growth not translate into nutritional improvements? Some left behind by income growth? Other priorities for expenditure - adult consumption, tobacco, alcohol)? Nutrition/health gender bias within the household? –But, there is a significant history of ignoring criticism of claims of female nutrition bias Harris-White and Watson; Jo Kynch; Gillespie & McNei Men are malnourished too – perhaps even more so (but anemia,..) –After the age of 35 (natality related mortality) male deaths exceed female »Dyson attributes this to fatigue related vulnerability to TB etc.
57 What to make of this? –Recall the 25 years of excess female child and foetus mortality in India because education and employment of women has not been working If educating and getting women into the labour force is not working, then what to do? –Recent legislation in India on dowry, domestic violence, sex-selection, midday meals, even NREGA.. –Dont base policy on poor analysis Goldacres 5 reasons [W]hy clever people believe stupid things (Ch 12, Bad Science): they –see patterns where there is only randomness –see cause where there is none (only correlation) –overvalue confirmatory information –seek out confirmatory information –assess quality of new evidence based on prior beliefs –Constructionist talk creating new languages for interventionist initiatives –Capabilities, functionings, opportunities, rights –Womens subordination, double/triple burden, endangered species.. Opportunities for experts and adepts -> institutionalisation –For example in the development industry
The development industry
59 To raise doubts about this claim is to challenge the morally unimpeachable… (slings and arrows …) To suggest possible reasons why it has become such an orthodoxy –Philosophy of science –Policy sciences –Social constructionism –Moral selving and the authoritarian turn The pious say that faith can do great things, and, as the gospel tells us, even move mountains. The reason is that faith breeds obstinacy. To have faith means simply to believe firmly - to deem almost a certainty, - things that are not reasonable; or, if they are reasonable, to believe them more firmly than reason warrants. A man of faith is stubborn in his beliefs; he goes his way, undaunted and resolute, disdaining hardship and danger, ready to suffer any extremity. Now, since the affairs of the world are subject to chance and to a thousand and one different accidents, there are many ways in which the passage of time may bring unexpected help to those who persevere in their obstinacy. And since this obstinacy is the product of faith, it is then said that faith can do great things. Francesco Guicciardini, Ricordi, Series C i (translated by Mario Domandi; quoted in John Dunn, 1972(?))