World Development 104 (2018) 344-357 Contents lists available at ScienceDirect World Development journal homepage: www.elsevier.com/locate/worlddey Stunting later in childhood and outcomes as a young adult: Evidence from India Rozana Himaz Deparment of counting. Finance and Economics, Oxford Brookes Univers Oxford OXS BP, UK ARTICLE INFO ‘ice ino ‘cepted 12 December 2017 ‘valle cline 30 Becerber 2017 eywois stunting ‘Adolescence Psychosocial Gender Ina ABSTRACT “This paper looks at pattems of growth faltering and catch up of around 1000 children as they moved from 8 to 19 years of age, fom middle childhood through adolescence co young adulthood, using Height for ‘Age Diflerence HAD) and the more conventional Height for age z-scores (HAZ) I also looks at what Vidual and household characteristics may have moved these children into or out of situations of nutsi- tional deprivation and how their stunting. profile in later childhood correlates with psychosocial ‘outcomes at age 19 and how itmay have intergenerational consequences. The paper uses 4 rounds oflon- iudinal data collected in 2002, 2006, 2009 and 2013 from Andhra Pradesh and Telengana, India when the children were aged 8, 12,15 and 19, The paper finds that there are significant gnder based bases in growth faltering later in childhood distavouring girs and that becoming newly stunted as an adolescent 's strongly corelated witha child reporting to have poorer relationships with peers compared to the group that were never stunted, We also find tha a girl experiencing stunting in middle childhood or ado lescence (even i they were not stunted at age 8 or eventually moved out of being stunted by age 19) cor relates significantly with offspring being shorter and thinner than the offspring of girls never stunted. “This is one of few, if any, studies that look at growth patterns in middle childhood and adolescence and outcomes asa young adult and the results are important fr their implications for further research and policy 1 Introduction ‘Stunting affects around a quarter to half of children in develop ing countries due to poverty, nutritional deprivation and burden of diseases (Grantham-McGregor et al, 2007; Victora, de Onis, Halal, Bldssner, & Shrimpton, 2010). Its consequences for outcomes later i life can be detrimental with significant negative effects on cog- nitive and non-cognitive development, schooling attainment and later outcomes as an adult in terms of earnings and productivity (Doyle, Harmon, Heckman, & Tremblay, 2009; Maluccio et al., 2008; Dercon & Sanchez, 2013, Hoddinott et al, 2013). Short sta- ture is also found to inerease the chances of giving birth to smaller babies and experiencing complications during pregnancy and childbirth (Black et al, 2008). Recent studies have also found evi- ‘dence that both contemporary and childhood health ofthe mother correlate positively with offspring health, and these effects are likely to be persistent (Bhalotra & Rawlings, 2011). Most papers ‘on stunting, however, focus on children under 5. This is unsurpris- ing given that stunting is argued to occur mainly within the frst few years of life (Martorell, Khan, & Schroeder, 1994). Moreover mal edd: sim @bronkes ace hiupsidiort0.1016i,workdey2017.12019 (0305. 750K)6 2017 Elsevier Al ight reser (© 2017 Elsevier Ld All rghts reserved. children usually enter middle childhood (defined here as ages 7— 12) and adotescence (between ages 13~19) with nutritional deficits accrued from eatlier on in life. However, children already stunted may ‘catch up’ later on childhood given appropriate conditions (Golden, 1994; Tanner, 1986) just as much as some of those who ‘did not enter middle childhood stunted may falter in their growth, and become stunted by the time they reach young adulthood, ‘There is little empirical research on growth patterns of those ‘who enter middle childhood stunted: Did they remain stunted as adults or move out of being stunted during adolescence? Were those not stunted in middle childhood falter in their growth during adolescence such that they were stunted as adults? What observ- able individual and household characteristics drave these results? ‘To what extent did vatiations in the stunting profile in middle childhood and adolescence influence outcomes as young, adults including cognitive and psychosocial outcomes, and health out- ‘comes ofthe offspring ofthe stunted children? This paper attempts, ‘to answer these questions using 4 rounds ofthe Young Lives longi- tudinal data collected for children from Andhra Pradesh and Telan- {gana from ages 7-19, There is some evidence in the economic and Public health literature, obtained using earlier rounds of the same Tongitudinal data set used in this paper, that growth catch up can‘Mina Word Development 1042018) 344-357, was ‘occur in middle childhood (Himaz, 2009) and early adolescence (Fink & Rockers, 2014; Outes & Porter, 2013). These studies use height for age 2-scores (HAZ) as the key unit of measurement 10 assess catch-up growth. However, a recent debate inthe literature argues that HAZ is inappropriate for measuring ‘catch up* in a lon- gitudinal dataset although it is appropriate for comparing groups of children between countries at a given point in time (Leroy, uel, Habicht, & Frongillo, 2015), The argument is based on the ‘observation that the HAZ is derived by using the difference between the actual height of the child in centimetres and the expected height according to the standard (HAD), divided by the age-sex based standard deviation for the reference population. This Standard deviation increases over time and is based on cross sec- tion data, Thus a gain in the HAZ value may partly be due ‘true gains’ in HAD (the numerator), but also due to the fact that the ‘denominator has risen even if the numerator has not. This means that observed ‘catch up’ when measured using longitudinal data isa statistical artefact if driven by increases in the denominator ‘that are higher than the numerator. Thus HAD arguably is a more suitable measure of assessing catch-up. There are significant dffer- ences in results depending on what you use. For example the ‘sub- stantial’ changes in catch up growth observed by Fink and Rockers (2014) that appear ‘equally likely’ in middle childhood and early adolescence are not quite substantial if one uses HAD. Instead, ‘atch up growth is more noticeable in adolescence. Thus this paper deviates from the previous Young Lives data based studies on growth catch up and faltering by (a) analysing the latest round of data for India (available at the time of writing) that includes information for the children aged 19, (b) using both HAD and HAZ to reassess faltering and catch-up effects in middle childhood and adolescence and (c) looking at differentials in ‘growth patterns between boys and girls in the sample (4) looking at how variations in the stunting profile between ages # and 19 have an impact on various psychosocial outcomes as a young adult as well as offspring outcomes. Psychosocial outcomes (Which refer to behavioural attributes ofthe individual) are measured using the rich data collected inthe Young Lives survey including measures of agency (the child's sense of freedom of choice to influence own life), self-esteem (overall evaluation of self-worth), self-efficacy (coping with daily hassles as well as adaptation after experiencing, all kinds of stressful life events), relationship with peers and par- ents, and general subjective wellbeing. The paper is organised as follows. Section 2 describes some ‘methodological issues and data used, Section 3 looks descriptively at growth patterns based on how the stunting profile changed among our children as a group as they moved from middle child- hood to adolescence and how using HAZ and HAD indicates diffe ences in growth patterns among boys and girs. Section 4 uses HAZ. to categorise the sample based on their stunting profile as those ‘who were never stunted, persistently stunted, moved out of being, stunted as an adolescent and moved into being stunted during ado- lescence, to glean insights as to what individual and household characteristics may have influenced height as an adult among. inl viduals in the different groups. Section 5 looks at how the stunting. profile correlates with psychosocial outcomes as young adults and the health outcomes of offspring for the subsample who became parents. Section 6 concludes. 2. Methodological issues and data 2.1. Reference values, growth faltering and catch up ‘The HAZ for a child proxies accumulated investments in child health and is derived by standardizing a chila’s height using the expected height and standaed deviation for a child of his (oF her) age and sex. The expected height and standard deviation come from the mean growth trajectory of a population of healthy chil- dren from birth to 19 years of age, as constructed by the World Health Organisation, referred to as the WHO reference 2007.’ The reference population mean growth trajectory is expected to be at the median of the growth standard. & population level deficit in height (calculated as the average of the individual height-for-age ifferences-HADs), is reflective of growth impairment caused by a deficient environment that may include poor diet, inadequate care and attitudes to health, a faced by the population of children under study. ‘A child is deemed ‘stunted’ if the HAZ is below -2 standard deviations of the mean, The -2Z-score cut offs used by the World Health Organisation (WHO) Global Database on Child Growth and ‘Malnutrition implying that 2.3% ofthe reference population will be classified as being stunted even ifthey have no growth impairment and are not unhealthy. Catch up growth can be defined as partial recovery froma linear growth deficit accumulated in the past. For recovery to happen children should grow faster than the expected velocity for their age and gender, making up for lost growth in height. But as CChrestani, Santos, Horta, Dumith, and de Oliveira Dode (2013) ‘observe in a systematic review of articles in the medical and public health literature as found in Medline/PubMed databases on catch up growth among children under 12, there is no uniformity in the operational definition of the concept of catch up. In recent {econometric literature such ‘catch-up’ has been identified by look- ing at the slope of the lagged HAZ in a dynamic model of nutri- tional status. But as discussed in the introduction, the use of HAZ to measure catch up when using longitudinal data, is debated. ‘Leroy etal (2015) suggest that when using longitudinal data, true catch up can be measured only using the HAD absolute values. HAD uses an expected growth trajectory based on a reference pop- ulation of children unlike HAZ changes that do not compare against an “expected HAZ trajectory”, Thus this paper avoids measuring ‘atch up growth using regression analysis and instead investigates it in descriptive terms using changes to HAD in Section 3, which is ‘compared to trends in HAZ. 22, Data “The data for much ofthis paper comes from the older cohort of the Young Lives longitudinal survey data for children, households and their communities collected in 2002, 2006, 2009 and 2013 from two regions in Andhra Pradesh (Coastal Andhra and Ray- alaseema), and Telangana, India, when the ‘index child’, was aged 8, 12, 15 and 19 years on average, respectively.” The original sam- ple contained 1000 children which dropped to 994 and 976 in the second and third rounds. By the fourth round the number was ‘951, Stl, the overall attrition rate of 4.8 per cent aver 11 years (averaging 04 per cent per year) is one of the lowest in longitudinal surveys ofthis nature (Barnett etal, 2012). ‘The sample is largely pro-poor, as the aim of the Young Lives project i to look at the causes and consequences of childhood pov- gpjiwneehain cdrom standardsfelgh fo agen). See also Onis (2007 for methods and deal pertaining tothe WHO 2007 standard. The ‘ung Lives Round «sample sein between ages 185 years and 12. yea In fer torte che childen ove 19 years of age the samp and usd sso ic the ats of these over 19 are founded dow Co 19, unde the assumption that roth vial eases because of epiphyseal fusion. typically ta skletal age of 13 years in pis and 17 years boys (loz Clark & Roemmich, 2000524 Tae 180) 4 se uty and ones 2014), Boyden (2014382015) and Boyden eta (2016) + See il (2008) Alderman, Berman, Kohls Nau, and Watkins (201) fr some examples of developing county longi datasets and thei atrtion aes ‘area bea igh 50 percent Some ese26 Hina Wotd Development 10 (2018) 344-357 erty. However, a careful analysis ofthe distribution of child charac teristics included in the sample suggests that the data covers a ‘wide variety of children in terms of wealth, consumption, similar to nationally representative datasets, Therefore, while not suited for simple monitoring of child outcome indicators (as the mean characteristics will be different), the Young Lives sample is appro- priate for analysing. correlates, causal relations and dynamics (Kumra, 2008). The Young Lives project also collected information for 2000 children over the 4 rounds from a younger cohort, aged 12 months on average in 2002. This data is used occasionally in this paper for descriptive purposes, as acknowledged where appropriate. ‘Appendix Table Al provides descriptive statistics (means, stan- {dard deviation and range) for the key variables used in this paper, as well as providing definitions for these variable. As the table shows, roughly half the sample of children are female. Overall, the average HAD across the four rounds indicate a worsening although not necessarily in a linear manner, moving from 8.93, in 2002 when the child was aged around 8 years to 11.72 at age 15 to ~10.57 as a young adult aged around 19. In contrast, ‘average HAZ seems to fluctuate around the ~1.56 standard devi tion mark that was noted in the first round. However, these aver: ages hide significant gender and other variations, as discussed in the analysis to come. 3. Growth faltering and catch up ‘Around 30 per cent of the children in our sample were stunted at age 8. However, less than half of these children remained stunted at age 19. As Fig. 1 shows, some children moved into being stunted later in childhood and adolescence while others moved out at various points, By age 19, around 10 per cent ofthe sample were persistently stunted since age 8, while 20 per cent were newly stunted in middle childhood or adolescence. So there seem to be considerable dynamics in terms of growth faltering between ages Band 19. “Table 1 provides some descriptive statistics that indicate how child health outcomes varied among our groups at age 19 com- pared to the reference group of children that experienced no stunt- ing.‘ The table shows that children who experienced some stunting in middle childhood remain significantly shorter than the reference ‘group even if they are no longer stunted. However, the Body Mass Index for age is significantly lower only for those children persis tently stunted and for those that were stunted in middle- childtiood but moved out of this category in adolescence. Those per~ sistently stunted also have significantly higher long term health, issues and a significantly lower perception about their own health ‘compared to those never stunted. Quite noticeably, a significantly higher proportion of those who moved into being stunted in middle childtood are girls (76%) as are those who moved into being stunted during adolescent years (62%). ‘Were patterns of growth different depending on the measure- ment used j.e,, HAZ or HAD? As Fig. 2 shows. conventional HAZ measures for both boys and girls show a fall between 12 to 60 ‘months. But there is an improvement in these averages between ‘ages 5 and 8, Patterns observed for boys and girls deviate from this, point onwards in our sample of children, The average HAZ seems to improve for boys relative to the reference population during mic- dle childhood (ages 8 and 12). Then there isa dip between ages 13, to 15 followed by improvements later in adolescence between *hges 13-19 hasbeen groupe as afloescenc forthe sake of clarity instead of breaking wp the group ar 13-15 and 5-19 a the latter pronder too mich information tts eter presencia format eter tana able Ina case ven the volt asscated ith growth daring puberty and adblescence the een tation nee netween cy snd iter adlecence nl neesaaly meaning Moved oit os ‘hen in Ne Moses sey Newly then in Fors a] fae Ney te SR) Stumcs E Bree12 stunted é since 15 Boas Baca Age of child Fig. 1 Stunting though mide eilahoad and adlescence aged 16 and 19. This ‘dip and rise’ of HAZ among boys may be indicative ofa slightly later onset of the pubertal growth spurt in ‘our sample compared t0 the WHO reference population. This is unsurprising given that age at puberty differs considerably between populations with later puberty occurring in populations ‘with a poor nutritional status, as is our sample which is drawn from children from poorer backgrounds. Unlike for boys, girls HAZ scores continue to deviate from the reference from age 8 ‘onwards until about age 15. The patterns emerging from the HAD analysis are different. Here growth deviations do not show ‘any improvement in middle childhood and continue falling wntit ‘age 12 oF s0 for both boys and girls, For boys, however, there seems to be some catching up occurring later in adolescence from age 15 ‘onwards. For girls, HAD seem to stop widening compared to the reference population in adolescence with marginal improvements ‘observable, as for boys, in later adolescence. Overall, growth falter- ing for girls seems worse than that for boys, after age 5, ‘One implication ofthe patterns in HAD and HAZ measurements js that the onset of puberty is perhaps slightly later for our sample ‘of poorer children giving rise toa higher growth velocity later on in adolescence compared to the reference population. The older ‘cohort data asks children directly about the onset of puberty. Roughly 27 per cent of the girls have started menstruating by ‘age 12. They have a mean HAZ of -0.93. Only 11 per cent of this, ‘group are stunted. Among boys 22 per cent have experienced a ‘deepening of voice and less than 3 per cent have any visible hair ‘growth on their chin by age 12. Just like the girls who have reached Puberty, a majority of these boys (95 per cent) are not stunted, ‘with an average HAZ of -1.05. In the reference population the ‘growth velocity for girls seem to peak around ages 11 to 12 and boys around 13 to 14, Thus it seems that the pubescent growth spurt occurs slightly later especially for children who are stunted in our sample than that in the reference population, similar to dis- ‘cussions in Kulin, Bwibo, Mutie, and Santner (1982) or Parent etal. (2003), Does this mean that the apparent ‘catch up' in the graph. above is simply the effect of later timing of the pubescent growth spurt between the sample and the reference group especially for stunted children rather than true catch up? This isa difficult ques- tion to answer but what can be noted is that although average HAD indicates a rise later in adolescence there are substantial differ- ‘ences among different groups of children, with some indicatingMina Word Development 1042018) 344-357, ‘able 1 w (Gender and health etcomes at 19 or eect groups of stunted versus never tuted eilren Never Stuned at 1219 Stunted a 12,1 Newly Stunted 319 Stunted a 8 oF 12 ‘Stunted (genstenly stunted) (moved to being stunted (enoved into being (Moved out of ing stunt {nmi childhood) Stued in adese) inna chodfadls) o@ o iis 04805 ost Heath Outcomes a age 19 round) Height forages score (HAZ) ost 264 2a Height fo age dflerence (HAD) cm 55182 1.02 Body Mass index forage nag 1a93" 58 ‘Own perception of gaveal helt 39375 as (very pour™ 1, poot=2. average 3, good 4, ‘very good) . Permanent sabi affecting workschool 00105 03 002 001 Nrof Obs at age 19 found 4) ao 3 103, 1 ‘ote: Anerits "and ™ mark that the mean 1) 38 te and 1p cent eel especie efor tha group was significantly dierent othe corresponding mean outcome the reerence group (ewe stunted a ro wma | eee Le tubo re 16 ig. 2. Mean height fo age 2 scores (HAZ) and height or ae iference (HAD) at ge 15,8, 12.1 and 19 yar for boys and sls separately Note: Data poms for ages and ae based onthe Young Lives young cohort round 1 and rund 2 data cole substantial faltering rather than catch up. To illustrate, Table 2 shows that the highest improvement of HAD in adolescence is amongst those who were stunted in middle childhood but moved fout during adolescence; 5.46 em for boys and 801 em for girs. ‘The second highest improvement is seen in those who were persis- tently stunted; 2.04 em for boys and 1.76 em for girls. So these are the groups that seem to have had some amount of growth ‘catch up’ But for those children who did not enter middle childhood stunted but went on to becoming stunted later on in childhood - roughly 15 per cent on the sample- growth faltered substantially between ages 12 and 19. For boys in this group, HAD widened by 10.92 em over adolescence while for gils it widened by 6.58 cm, In contrast HAD widened by less than 1 cm during the adolescent yeats for boys and girls who had never been stunted. Overall, therefore, the improvement in average HAD deficits in adolescence is driven by those that moved out of being stunted and those persistently stunted (32 per cent ofthe sample in total, reflective perhaps of delayed pubertal growth spurts. But the aver- age improvement conceals the fact that for around 15 per cent of ‘the sample substantial growth faltering has occurred, tein 2002 ana 206, fo 200 cilren. The est othe ata points are bse on the Oe 4, What drove growth dynamics in adolescence? ‘Himaz (2009) argues that even ifa child starts middle childhood ‘with significant shortfalls in height accrued from earlier on in life, ‘nutritional interventions and adult female education may have a positive impact on linear growth and perhaps mitigate conse- quences of early age stunting by the time the child reaches age 12, using Young Lives data for 2002 and 2006, That analysis is extended here using further rounds of data collected in 2009 and 2013 to ascertain what individual and household characteristics ‘may be helping children move into stunting and out of stunting in adolescence. The model estimated is specified as follows: Ya = DXi + Badin + Baler +E a where y refers to the stunting profile of child {at time t (when the child isa young adult aged 19) with y= 1 ifthe child had never been stunted (ie. not stunted at age 8, 12 or 19),y=2 ifthe child moved ‘out of being stunted during adolescence (ie, stunted at age 8 or 12 Dbut not stunted at 19 indicating stunting in middle childhood butus able? atch up o fuer ering of growth in adlescence? Hina Wotd Development 10 (2018) 344-357 ‘Disuareated Sub Sle Never Stunted ersistentystnted Boys MAD in roune 1 (2) HAD nro (6) HAD in round (0) {Change inFIAD apes 8-12: (Da) (Change in HAD ages 12:19 (2.0) Change in AD age 8-12: (0)(a) 083 a Change inHAD ages 12-19:(¢h(0) a4 1768 Number of aeerations al Sample ‘Moved ino being tune nmi. chhodados Moved oot of bing stunted inmidehhood ads moving out during adolescent ages 13-19)-y~3 if the chilel moved into being stunted during adolescence (Le, not stunted at age 8 and 12 but is stunted by age 19) and y =4 if the child has been persis- tently stunted through middle childhood, adolescence and is stunted as a young adult (Le. stunted at ages 8, 12 and 19); Xis a vector of child-speeific fixed characteristics such as gender, birth ‘order, caste (Scheduled Tribe, Scheduled Caste, Backward Caste ‘with Other omitted), mother’s height in centimetres (log), and a proxy for the quality ofthe chile’s diet; 8 is a vector of household ‘characteristics for child fin house hold j,at time period ¢ ~ 1, which refers to the time when the child was 12 years old at the start of| adolescence (round 2). The household characteristics include the ‘wealth index, mother’s education, demographic composition of children in the household by age and gender (proportion of females in the household aged 0-5, 6-12, 13-17, 18 + and males aged 0-5, 6-12, 13-17, 18+ with females 18+ omitted) region of residence (Coastal Andhra, Rayalaseema, Telangana with Telangana omitted), and the area of residence (rural or urban). Finally, 0 refers to com- ‘munity level chatactersties such as whether the sewerage disposal facilites are considered ‘good’ as opposed to being bad or ‘50-50 and if access to electricity was considered to be good, at round 2. ‘These are both dichotomous variables. “The inclusion of these control variables have been based mainly ‘on the theoretical and empirical literature based on child health in India, For example, gender, birth order and demographic composi- tion of household are included as there is evidence to suggest that household resource allocation and attitudes to investment in health may depend on these factors (Behrman (1988), Jayachandran_ and Kuziemko (2011), Jayachandran and Pande (2015), Pande (2003), while mother’s height is meant to capture 4s far as possible the influence of genetic factors especially during, adolescent years (Acido et al, 2013). Caste is a proxy for access t0 health related services as well as attitudes and practices related ‘to health inputs.” In order to account for the possibilty that dict and food security may have an impact on adolescent linear growth, ind caste sytem s form of scl stratification thats rund 2000 ears. e dvdes nds inc igi Near groups based on thir karma and dharma, ‘which mean atk and dy respctivey At the top othe racy ae the teaches fd itl (Brain) folowed by warts and rulers (Shatias} Wades and Ione lenders (Vass) and a the bottom thor who di menial obs (boda). The main cases are frter divided into around 25/00 occupation specie sub castes Outside the tational castes vere the Dalits or untoehabes (sched {ste} undertaking dhe fowest ofthe menial jobs. The ate system dictated alos very aspect of Hind religions and socal efor many centre, URE Upp ‘tes ewe oer eaten The Indian conection banned este baed hriminston in 1947 while wider acess toecation and social modity have reduced somewhat the influence of astes: Howes. sgitcant ease based derences si ean in sais spheres (rocah, 200: Deshpande, 2000} (as argued in Belachew et al, 2013; Rogol et al, 2000) I ereatea diet ary diversity index based on Swindale and Bilinsky (2006), The index. is mainly a proxy forthe diversity ina chile’ diet in terms ofthe var- ous food groups consumed, but also food security.” The community level variable regarding access to good sewerage facilities is included as there is evidence to suggest that sanitation can play a strong role in child health (Spears, 2013) while access to electricity proxies the level of ‘development’ in the community the child comes from, and thereby accounting, atleast partially, for any community-level fac- tors that maybe affecting adolescent health, The region of residence 72 000 ‘Notes: Al egresions include the flowing variables apart fom those reported: ith order household sie, boushold demographic composition a eden ocd esto elect a seteraze disposal falties athe community level Robust standard toes (adjusted for 10 community lusts) npaentheses "p< 1,” B05 Pel composition, with better quality cheaper food such as coarse grain substituted for poorer quality often more expensive refined food (Deaton & Dréze, 2009:56-58) or due to the increase in spending, ‘0n non-food items (Basole & Basu, 2015)" The signs for other control variables are as expected, although not statistically significant. ‘The results from column(2) shows that the most significant fac- tor that is associated with moving a child into being stunted during adolescence is gender: the relative log odds of moving into being stunted in adolescence compared to never being stunted falls by (0.57 when moving from being female to mal. Similarly, belonging, to the Scheduled Caste as opposed to Other, positively affects mov- ing into being stunted compared to never being stunted. In con- trast, increased dietary diversity and wealth both reduce the relative log odds of moving into being stunted compared to never being stunted. Thus being female, having a diet that is not varied, household level wealth deprivation and belonging tothe scheduled caste are all factors that significantly influence moving into being, stunted in adolescence. Column 3 shows that the relative log odds of being persistently stunted compared to never being stunted is increased when com- ing from Scheduled Caste, Backward Castes or Scheduled Tribes as ‘opposed to Other castes. The strong correlation persistent stunting. hhas with certain castes suggest that attitudes to health and nutti- tion, health inputs as well as access to services may affect invest- ments in linear growth in adolescence significantly. Persistent stunting compared to never being stunted is associated negatively by mother’s education and positively with coming from Ray- alaseema as opposed to Telangana. The Wald test at the bottom "The pusze a to why cost section data may indicate postive colton becweenciorie consumption an income But why thre i 3 neatve corelton ‘en Tooking dats eve ime fn India scsed more broly in Ds rds 2005) ‘of the table tells us that our model as a whole fits significantly bet- ter than an empty model (Le, a model with no predictors)” Column 4 shows thatthe key characteristics at age 12 that cor- relate significantly positively a child's final linear growth outcome are being male, genetic effects, dietary diversity, mother’s educa- tion, household wealth and being from ‘other’ caste. ‘The robustness of these results are checked for by changing the specification to be more parsimonious. I exclude mother's height (ince this value was not available for 43 observations and thus limited the number of observations | used in the previous estima- tion) and household demographic composition, All the results dis- ccussed previously remain robust to this change in specification (Appendix Table A2)."" It is acknowledged here that although the issue of reverse causality is minimal under the specifications above since the explanatory variables are lagged, it does not preclude the possibilty that there may be unobservables that may have not been ‘captured in the specifications (inspite of the efforts to include suit- Vals conduct a Wald test to sc if the dependent categories weenie (1) reve stunted, 2) moved out (3noved in and (4}persistent can be combined to ‘rovide mace etiient estimates. The tests that none ofthe pls fates are Insingshabe wth eset to variables inthe models atthe per ent eve. THUS there illest! support to combine any of the dependent categories Hwsman tess of Independence of inelevane Alternatives (A) assumption. a Stringent assumption of the model instep signa evidence spans Ho ‘ds{ Outcome! vs Outcome K) are independent a ther alternatives) ° Diagnostic tests done onthe fll model versus nested dl, bse on Long an reese (2014) inate thatthe Bfleence In Bayesian information criterion (I) econ the models prvisessopor forthe more parsmonious model. Te Ic st {ends to favor more parsimonious modes when snp size pts le: However, an sleratve rte sch a the diferecein chi-squared stasis between pasino350 Hina Wotd Development 10 (2018) 344-357 able proxies)-This means the relationships l estimate ae largely cor- relations rather than being causal 5. Stunting profile and outcomes 511. Psychosocial outcomes In this section 1 investigate how changes in the stunting profile in middle childhood and adolescence are associated with cognitive and psychosocial outcomes as a young adult of age 19. This focus is different to that of most papers in the empirical literature that look at stunting in early childhood and subsequent psychosocial out- ‘comes. For example, poor early investment in Health is hypothe- sised to have significant negative impacts on psychosocial ‘outcomes later on Chang, Walker, Grantham-McGregor, and Powell (2002), Walker, Chang, Powel, Simonoff, and Grantham- McGregor (2007). Supporting this, Dercon and Sanchez (2013) find ‘empirically a correlation between height-for-age at age 7-8 (a proxy for early investments in child health) and psychosocial com- petences (‘non-cognitive’ skills such as school aspiration, self ‘esteem and self efficacy) at age 11-12. They stipulate that this pro- vides evidence for an underlying relationship between undernutri- tion earlier on in childhood and the formation of non-cognitive skills. But psychosocial skills ~especially later on in childhood ‘can also be shaped by height relative to peers rather than by long-term effects of early investments in nutrition. For example, shot statute may be correlated with increased experience of being teased, juvanilistion (ie, socialisation that can happen according to height-age rather than chronological age) or overprotection by parents as stunted children are (or are perceived to be) physically ‘weaker than peers that may itself cause victimisation by peers, all contributing to poorer psychosocial outcomes (Stabler etal, 1994; Steinhausen, Dorr, & Kannenberg, 2000). Although compelling ‘especially in clinical data based studies, this view is not always supported empirically in non-clinical settings (Voss, 2001; Voss & Sandberg, 2004), For example, the longitudinal community-based ‘Wessex Growth Study compared psychosocial outcomes for a sam: ple of very short (less than the 3rd centile) but otherwise healthy children at school entry (around age 5) with those of controls with average height, at ages 7-8, 11-12 and 18-20 to find no significant differences in terms of Adolescent to Adult Personality Functioning ‘Assessment (ADAPFA) scores between the groups (Downie, Mulligan, Stratford, Betts, & Voss, 1997; Ulph, Betts, Mulligan, & ‘Stratford, 2004), The ADAPFA measures functioning in the domains ‘of education and employment, love relationships, friendships, cop ing, social contacts, and negotiations. Thus the empirical evidence found in the literature on whether stunting in early childhood has ‘a causal link to psychosocial competencies -via the nutrition ink or the relative height link: is both sparse and not particularly strong in terms of evidence, But almost no work has been done on what impact becoming stunted later in childhood has on psychosocial ‘outcomes and this is what we turn to next. ‘The model | estimate is as follows: Yq = T+ B+ Badisn a + Bae + 2 here y refers to the psychosocial outcome of child fat age 19 (time 1), which isa continuous variable reflecting standardized values for ‘measured psychosocial outcomes, These outcomes are agency (Le. ower or ability to influence one’s life and related to the concept ‘of “locus of control’ as discussed in Rotter (1966), self-esteem (ie, an evaluation of self worth which follows an adapted version of the self-esteem Scale in Rosenberg 1965, self efficacy (coping ‘with daily hassles as well as adaptation after experiencing all kinds of stressful life events, see Schwarzer & Jerusalem, 2010) and rela- tionships with peers, parents and general self perception as effec- tive, capable individuals taken from the Self Description ‘Questionnaires as discussed in Marsh and O'Neill (1984). The full list of survey questions included to compute these scales, tailored to ‘suit a context of child poverty and specific dimensions ofthe child's living circumstances such as housing, clothing, work, school, are reported in Table-A3 in the Appendix. Vector X refers to child specific ‘characteristics such as stunting profile (i.e, five categorical variables as described previously, with never stunted being the omitted ‘group), gender, birth order, caste: vector 6 reflects household charac- teristics for child i in house hold j, at time period t— 1 (when the ‘child was 8 years old). The household characteristics include house~ hold size, wealth, father’s and mother’s education in years, region of residence. area of residence (ruraljurban).Finaly, 0 refers to commu- nity level fixed effects used to control for unobserved heterogeneity among the communities and ¢ isthe random errar term, Eq. (2) 8 ‘estimated using OLS regression analysis. ‘The results reported in Table 4, columns 3 and 4 show that mov- ing into the “stunted” category in adolescence (between ages 12 and 19) is strongly correlated with a child reporting to have poorer relationships with peers (at the 1 percent level of significance) and lower sef efficacy (at the 10 per cent level) compared to the group that were never stunted. Our previous discussion suggests that this ‘group of children seem to be ones that experienced substantial ‘growth faltering between ages 12 and 19. Apart from this, persis- ‘ent stunting i positively significantly correlated at the 10 per cent level with having strong relationships with parents. What these correlations suggest that i there is no support for the hypothesis, that early inadequacies in health inputs have long term negative implications on psychosocial outcomes as a young adult. Iit did hhave an impact, then the profiles where the children were stunted at age 8 (such as persistently stunted) should have shown a strong, negative correlation between being stunted at age 8 (a proxy for poor inputs in child health in early childhood) and psychosocial ‘competencies as we measure them." We will return to discussing ‘the significant correlations we do find shortly, after noting several ‘other significant explanatory vatiables in the results. Gender, for stance is significant, with being male exerting a positive effect ‘on agency and self efficacy but a negative impact on most other mea- ‘sures of psychosocial wellbeing compared to being female. The father's and mother’s education is significant for agency and se eff- ‘cacy witle household wealth i significantly positive for agency. self ‘efficacy and peer relationships. Caste does not havea systematic sig- nificant influence on outcomes, but coming from Coastal Andhra and Rayalaseema seem to exert a positive impact compared on all out- ‘comes apart from relationship with parents, compared to coming from Telengana (the omitted category). A higher birth order signif- ‘cantly increases a child's general self perception as an effective, cap- able individual at age 19, possibly due to the responsibilities and respect culturally placed upon an older sibling (Cicvell, 1994). Returning to the results regarding stunting profile and psy- ‘chosocial outcomes, why would growth faltering in adolescence {among children not previously stunted) have significant impacts ‘on peer relationships and self efficacy while becoming stunted at ‘other ages, including persistent stunting, do not? One explanation ‘may lie in the relative height hypothesis that being shorter in sta- ture compared to others ofthe same age imposes significant chal- lenges. Were these children who moved into being stunted in adolescence bullied or teased more by peers, for instance? Descrip- TY See Deion at Khan (20%) foe moe detas on the development of the agency and se eter messes using Young Lites dal or the oer messes well as a discussion abou valde. sus face in pling these instruments and ‘leat ae Yorke and Orgando Perla fenton) "To check the robustness of this elim we ronan alternative specification of (2) ‘xcaing the suming profile dummies but icing instead HAZ 3 abe 8. The ‘rable ot signs anya he regressions (unepated)‘Mina Word Development 1042018) 344-357, ssi ‘atte 4 ‘Suntng prot, characterises a age and psychosocial outcomes 2 a young aul aged 18. Dependent variable ‘ney Sa ete Seifeccy ee retonshin Parent eaionship “General a a a wo et © ‘Stunting prof (reference group: never stunted) "Moved in sdbiesence ‘007 oo ans 025" 05 as (oon cout) (07) (0.08) (0.08) (0.06) (008) (10) (a1) (008) (0.08) (07) (006) (008 (oo) (008) (08) (006) Moved aut sid ehood 06 03 ‘01 ‘0! 05 oor adobe. (09, (0.06), (0.05) (005), (07), (oo) (ender (i= male) 20" 025" ous” 03" a3" act (004) (006) (08) (aot) (006) (oot) faint onder 00, 00 oor 01 00 03" (oon (0.02) (102) 02) (0.02) (oon) Father education for 001 “oot “00 “too ‘00 (oon (oor oot 000) (oo) (000) Mothers education or 001 02 01 01 or (oon (oon) coor coon (oor) (oon Westin, 2002 fe 024 oar ar 026 0 (013) (021) (16) 019) (oun, (013) Scheduled Caste oo 005 02 ‘08 ae ‘01 (00s) (ory (oon) (0s) (008) (07) (oon) (012) (oon) ox) corny (010) ackwara caste ‘om 02 000 ‘0s 01 ‘ane (006) (00) (0065) (005) (006) (006) (013, (021) 015) (as), (028) (ary (036 (038) (010), (040) (012) aylascema 10 135" as ono 03 (004 (008), (0.05) (0.06) (005), constant 0" “02 “oss ts “tno wal" (019) (023) com) (a9) (029) (on Dbseratons mn op on m2 92 32 Resquared 025 on 027 019 020 020 [Rs Al pesos alta contol for uscd size and connmunty xe elles: Rabat tandard eros adjusted for 100 community uses) parehesea™ p< ON, “peas. ped ‘ables Eiprience of eng ule, reatment by adults and sled crcumstaces during adolescence" [Never Moved into being stunted Moved into being stunted in Moved out of ing stunted Pasistenthy unted in adolescence ‘idle cildood Inadalescence Stunted a) o “ 6 Tapeviece of Being baled by pews bese (proportion) . Otter types 24a, 025, 2 030, ‘Adults treat child a5 well as they do thers of 090082 33” os 080" same age (proportion) Marie (proportion, 2013, ars 30) 019, ors on Enoled asthe (proportion), 2013 055032! oe 030, a6 ‘ete for paymtoney (hows day) 2013 721.20 135" 1a" 10 Dbeenatons| “01a 3 13 1 Mian outcome fo the espectve values i calm ¥ compared wilh oresponding ean values i columns 05, oes for Salsically signin dfleencs "10 pet 2 Se able Rt for deals on how these indices are calculated, > sed on round survey question Als in my community teat mea well as they do ater ciden at my age Experience coves 1M eespanse was gre or tong agee, otherwise tive statistics provided in Table 5 show that children who experi ence stunting in childhood, regardless of when they became stunted, report higher incidence of being physically bullied by peers in ado- lescence (between ages 13 and 15) compared to those never stunted, Table A4 in the appendix describes how the indices for bullying are constructed, Scunted children also feel they are treated less well by adults than their peers. But once controlled for con- founding factors in the baseline analysis, children who were newly stunted in adolescence indicate a strong correlation between their stunting profile and psychosocial outcomes such as relationships with peers and self efficacy, unlike other categories of stunted children.ss Hina Wotd Development 10 (2018) 344-357 Could this be because of omitted variable bias since unobserved determinants of a child's psychosocial development may also be correlated with what may have driven a child into being stunted in adolescence? Such determinants could be, for example, tunobserved parental of household inputs. Such inputs and atti- tudes may be shaped in part by poverty, which we account for some what imperfectly in our baseline specification. But it could also be because children in this group who are largely female (62%), experience different contemporaneous life circumstances to others that affect their relationship with peers in particular and self efficacy. They could have married, for instance, and expe- rienced in consequence less mobility and freedoms in the homes of their husbands’ affecting these outcomes. Or the boys and girls may have different time use patterns due to having dropped out ‘of school or spending more time in paid work, both instances that presumably reduce opportunities for positive experiences in socialising with peers and adults. For example Pells, Portela, and Revollo (2016:34) quote from the young lives qualitative surveys to argue that caregivers and teachers describe children out of school as being undisciplined and a bad influence on those that attend school and so, encouraged to be disassociated by those who attend school. The latter rows of Table 5 compare the propor- tions married, enrolled at school and time spent on income earning activity among children by stunting profil. For almost all these variables, apart from ‘married’, the differences between the aver- age values for all the stunted groups are statistically significantly different to the non-stunted group. For the vatiable ‘married’ df ferences are significant only among those that became stunted in adolescence and those never stunted. Thus 30 per cent of those who experienced growth faltering in adolescence are married com- pared to around 17 per cent inthe other groups. It is possible then that ‘married is corvelated to some unobservable (such as parental attitudes perhaps) that affects this group but not other groups of children and | therefore reestimate equation (2) including married as an explanatory variable. The results reported in Table AS panel ‘A show that being married does exert a significant negative impact, ‘on agency, relationships with peers and parents but does not change the previously found significant association between becoming. stunted during adolescence and peer relationships. ‘There is no longer a statistically significant association between being newly stunted and self efficacy at age 19. ‘The baseline specification is then re-estimated with further controls that indicate other contemporaneous circumstances that might have impacts on psychosocial competencies. One such vari- able is when the child is currently attending school, Another is con- ‘temporaneous household wealth and controls for household vulnerability to a variety of exogenous economic shocks (natural disasters such as droughts, floods, etc, and income shocks arising ‘out of job loss, stolen crop, etc) that occurred during adolescence ‘We also include the child's BMI at age 19. Finally we include prox- les fora chila’s psychosocial outcomes at age 8, where available, 0 remove as far as possible, child specific fixed effets. The results are shown in Table AS panel B, As with the previous robustness check, becoming stunted during adolescence remains significantly corre- lated with peer relationships but not self effcacy."? ‘The baseline results corroborated by the robustness checks sug- .gestthat there isa relationship between becoming newly stunted 438 an adolescent and peer relationships at age 19. One reason for this could be that children in this group are more sensitive to the negative effects of short-stature (such as bullying and differential pn fe veh the result fom this extended obsess check hat the spectetian mays rom revere casalty acre BM a hows wealth ‘ould affect prychoroca outcomes and vice versa Thus the results indicate ‘arelton rather than causiy Reverse causality Ves ofan ss nthe sine ‘specication lagged explanatory varles ar sed treatment) as they have had less time to build adequate coping strategies, unlike those persistently stunted or those who became stunted in middle childhood. Moreover, the baseline results also showed that those persistent stunted enjoy better relationships with their parents, possibly arising out of parents being more pro- fective towards these children who on average have more long ‘term health issues and significantly lower BMI forage (see Table 1). ‘These closer relationship with parents can help reduce sensitivity to negative peer behaviour and improved resilience. Several stuc- ies support this contention including Dekovié and Meus (1997) who argue that the quality of the parent-child relationship influ- fences an adolescent's self-concept which in tum affects his or her integration into the world of peers. 52. Ofsprng health outcomes | now move onto looking at how changes in the stunting profile in middle childhood and adolescence are associated with the out- ‘comes associated with the offspring of these children. Out of the ‘951 children in our round 4 sample, 107 (101 girls and 6 boys) ‘went onto have children themselves. In order to ensure compara- bility with the literature that focuses mainly on the transmission ‘of poor health from mothers to their offspring, only offspring born ‘to female index children are ineluded in the discussion and analy sis below. As shown in Appendix Table AG, the 101 girls in our sam- ple who had offspring by age 19 were slightly older on average than those girls who did not (105 years as opposed to 18.98), with a higher proportion (44% as opposed to 24%) reporting to have reached menarche at the time of round 2 (age 12). The girls who had offspring were statistically significantly taller than those ‘who did not at ages 8 and 12, but these differences reduced over time such that by age 19, there was no statistically significant dif- ferences in height. Although the rough proxies we have used for health outcomes between the two groups are not significantly dif- ferent, girls who already have offspring came from predominantly rural areas, from households that were significantly poorer i all four rounds and had attained levels of formal education that were significantly lower than those who did not have children. There ‘were no differences in the representation of castes between the two groups. ‘Out of the 101 girls who went on to have offspring by age 19, twenty-three had two children each, and thus the total number ‘of offspring in the sample is 124. However, 6 of the offspring had died by the time round 4 interviews occurred, Also excluded are few observations, where appropriate, if data is. missing or recorded values are outside the reasonable range (for example HAZ is outside range +6, WAZ outside range ~6 and +5). In Table 6 columns (1)-(5) below, I look at how the stunting, profile of the Young Lives children (reduced to just a dichotomous variable that equals one if a child experienced some stunting or persistent stunting at ages 8, 12 and 19 and zero if the child expe- rienced no stunting), offspring age, gender and household char teristics in round 4, namely, wealth, region and rural residence, correlate with offspring HAZ, weight for age z-scores (WAZ), inci dence of offspring still being alive, pre-term birth and the mother hhaving had a difficult labour. ‘The sample size is small and the results are indicative rather than being definitive. Moreover, as noted earlier, the sub sample of girls who are already mothers is selective, in that they come from relatively more disadvantaged backgrounds in terms of ‘wealth, for example, although not necessarily more disadvantaged in terms of accumulated investments in health as captured by HAZ ‘r stunting profile. Nonetheless, the fact thatthe group is selective in terms of wealth suggests that there could be unobservable such as attitudes to health that are correlated with the mother’s stunt- ing profile as well asthe error term, that may bias the coefficient‘Mina Word Development 1042018) 344-357, sss ‘atte 6 spring outcomes regressed gas pareat Suan pole Dependent Varable HAZ WAZ ——Sil/Alve Bomeary Difficult Labour HAZ WAZ Still Alive Borneaily_ifcu abour (novel o o oO oO o (9) Stunting roe ‘somes 095" a7!" -00e" 0s a0 (043) (027) (as) (007) «at0} . ' (os) (029) (aor), (a6) (a0) essen tuning zr 04-018" 004s : : 075) 057), as) (as) aa) otsing age aw” ais a9" at? (012) (a0) (a2) (ooo Otprng gender 0316s oma 038 01k on) (033) (024) (a2), (007) (00) 035) (025) (a2) (a6) (09) Weatt index 2012 200275" = “anys 20726" “907 aed (12) (09) ao) (02) 033) (125) (100) (as) (024) 033) coastal Andhra = 0380} ogra “ta ox) lor “900 aan “as (066) (02) (aus) (a2) «2} (66) (02) (0or) (a) «a2} aylascema aos 013002 ‘001 as” ‘1 013” ‘001 033" (045) (028) (age) (08) ade) (04s) (028) (02) (ans) ua) oral 49 “024 002) “004-008 #9) "025 0m 003.0 (oar) (031), (002) (008) (a2) (047), (032), (002) (006) (a2) Constant 292" 294" 206" 225” (195) (oan) 02) (079) ‘oie For ess nc (1) {a)reterence category for some nting nd perstent suntng ener stunted Regresson= (1,2) (6) and (7) areO1S estimations while the rest are probit estimations with marginal eet eporte.. Sanda errs adjusted fr custers in same mothe. fon the stunting profile. Unless appropriately corrected for this sample selection bias the results produce at best some indication of correlation rather than causation, Bearing these caveats in mind it can be noticed from the regression results that there seems to be evidence ofthe intergenerational transfer of poor health outcomes, ‘with those girls that experienced stunting in middle childhood or adolescence more likely to have children with lower HAZ, lower WAZ and be more likely to have offspring who died in infancy. In ‘order to glean some insight as to whether these results are driven ‘mainly by mothers who were persistently stunted, columns (6) 10 (10) present results where the stunting profile is split into three categories rather than two: those persistently stunted (Le. stunted at age 8, 12 and 19), those that experienced some stunting (1c, stunted at age 8 or age 12 of age 19 but not persistently) and those that experienced no stunting at all (reference category). The results indicate that persistent stunting in mothers correlates significantly with offspring mortality (column 8). Mothers who experienced some (but not persistent) stunting in middle childhood and adoles- cence seem to have offspring who are significantly shorter and thinner than mothers never stunted (columns 6 and 7). It is not clear why those persistently stunted do not have surviving chil- «dren with poorer health outcomes compared to the never stunted ‘mothers, but itis likely that the fittest survived and are possibly have health inputs that compensate for the loss of thei siblings. ‘The latter is of course conjecture that needs further investigation using the qualitative data the Young Lives project collects roring patterns for the parents, colurnns (1), (2) (6) and (7) also indicate that growth faltering seems to happen with the off- spring as well, with age having a significant negative impact on both WAZ and HAZ. Moreover, sharp deviations to HAD starts within the first few years of life for offspring, similar to trends ‘observed for index children in Fig. 2, With regard to the other con- ‘rol variables used, it can be seen that household wealth and living, in Coastal Andhra compared to Telangana exerta positive influence ‘on WAZ, Column (4),(5),(9) and (10) indicate that stunting profile does not have a significant effect on premature birth or having a dlifficue labour. peal ped ped 6. Conclusions ‘This paper used longitudinal data from four rounds of the Young, Lives older cohort survey for Andhra Pradesh and Telangana, India, to investigate how growth faltering beyond early childhood can have an impact on outcomes as young adults and of the offspring, ff these young adults. One of the first things noted was that although around 30 per cent of the children in our sample enter ‘middle childhood stunted, there is considerable udity in the chil ‘en's stunting profile as they move through middle childhood and adolescence to reach young adulthood. Adolescence in particular is 2 period where children -especially boys- who were previously stunted seem to catch up in terms of growth, at east partially, and move out of being stunted. More than half the individuals ‘who are stunted at age 19 are those that moved into being stunted during middle childhood and adolescence. The incidence is much higher among gitls, possibly linked to inadequate nutrition, wealth deprivation and factors associated with attitudes. It could also be related to health practices, given that caste is associated signifi cantly with moving into being stunted, Ths is an area that requires further investigation, Persistency in stunting through childhood and adolescence was positively correlated with mother’s education level as well a caste Growth faltering and catch-up patterns are different depending ‘on whether height for age difference (HAD) or the more conven tional (HAZ) are used, When HAZ is used as the measure of growth changes, it shows that for boys, the average HAZ improves relative to the reference population mainly in mide childhood and then again later in adolescence. For girls the trend is a continued fall in average HAZ from middle childhood onwards until about age 15, However, when HAD is used, growth deviations do not show any improvement in middle childhood for either boys or girls Instead, it continues falling throughout middle childhood until age 12 or so with the deviation from the heights ofthe reference population higher for girls than boys. After this point, HAD stops ‘widening for girls with marginal improvements observable later adolescence. For boys, HAD widens during the early adolescentssa Hina Wotd Development 10 (2018) 344-357 ‘years before improving late in adolescence. Overall trends in HAD Show that growth faltering in girls is higher than that for boys, after age 5, particularly during middle-childhood and pre- adolescent years. In the text it was argued that some but not all, ‘of these paiterns in growth may be explained pethaps by delayed pubertal growth spurts among children in our sample compared to ‘those of the reference population, ‘The results also show that moving into the ‘stunted’ category in adolescence (between ages 13 and 19) s strongly correlated with a Child reporting to have poorer relationships with peers compared to the group that were never stunted, This may be due to these children being more sensitive and less resilient to the effects of peer bullying and other difficulties arising out of short relative sta- ture as they have had less time to build effective coping strategies unlike those persistently stunted. The results for those newly stunted also do not indicate significantly positive relationships ‘with parents unlike those persistently stunted, which also reduces their resilience to the negative effects associated with short sta- ture. Psychosocial outcomes at age 19 were also significantly influ- ‘enced by factors such asa child's gender, parental education levels, household wealth at age 8, birth order and region of residence. Interestingly, there is no evidence that poor investments in early nutrition (as proxied by being stunted at age 8) has had long, term negative impacts on non-cognitive outcomes at age 19. ‘Although the sample of offspring is small and therefore results tenuous, there seems to be some evidence of the intergenerational ‘able a1 transfer of poor investments in health between mothers and their ‘offspring, as those who experience some or persistent stunting in ‘middle childhoodjadolescence have offspring who are significantly shorter and thinner than those who were never stunted as well as having a higher chance of their offspring dying in infancy. The strong correlation between offspring mortality and stunting seems {0 be driven by mothers who were persistently stunted. So growth, fluctuations during the pre-adolescent and adolescent years ~espe- cially if such fluctuations cause growth faltering and stunting in a girl child- is associated significantly with offspring health and mortality outcomes. This area requires further research and can possibly shed some light on how to reduce the inter-generational persistence of poor health outcomes. For example, stronger evi- dence in this area, perhaps through the use of further rounds of ‘Young Lives data that will become available in the future, can lend support to policy recommendations that target female nutrition «during middle childhood and adolescence to reduce the intergen- ‘rational persistence of stunting. Other policy recommendations ‘an include family planning, and if pregnancy does occur during teenage years, maternal and child health interventions. ‘Appendix A ‘Simmary statics for sled aril for indo chin present in al ands ofthe suey Varabe Obs_Mean ‘Skew. Min Max ——_Vanale Description ‘index hd arate No 2002 ea san 26211175 HAD: Height for Age Dierence calculated as the diflrence between Han 2006 m3 lost “4424 1798 the actual height ofthe cid in centimetres andthe expected Regt HAD 2009 So ana 7a “3479 162° according the WHO 2007 standard Hao 2013, 740s? 631 278 88s xz 2002 $0 158108 4512 HAZ: Height or Age sore aeuated as HAD divide bythe ae sex sex sts, ° 1 Soc id= 1 mae 0 female Bie oder Se ' 13 firth order ofthe index child compared eo siblings bor tothe same ‘mother who survived fo more tan 24h ater bith Dicey diversity, 2009 951473 ° 7 Index can ange fom O (last verse) to ost verse) counts the numberof fot ites in the following groups consumed by the index Inthe last 24: () grains. rots tubers i amin Aric vegetables frit and eran meat i) eer fis and vegetables meat on nan) an fish (veg: (pes ad legumes (i) maka dary products: food cooked toi at. Method follows Sinale and binky (2006) Housed Characters Mothers eduction 94628 as ° 14 Highest choo grade obtained exluing pre-primary futtersedocston 847 4A ° 14 Highest ehool-grate abained excluding re pray Wialthindes,2002 951 OAT. at 01 03nd anges ro (least wealthy) tT (most wealthy) constructed ‘ing te simple average of 3 sub ines fo nosing quality erowalng enter we ta ld ot wal apd No) acces to services (lect, safe drinking wate, sartton and adequate fe) and onsumer durables (owns at least one ase ut of 9 such a TV. Dlejele} Se spin 2014), pa for further deta others Height 96 so a as 51 Matera eight fo the nearest 0.1 em logs Howsebold size 2002 «$2082 04 2 32 Thetotal number of members nthe households st soz) at ° 1 The asta cl belongs to: Scheduled cst (SC) Backward caste), st ss ano aad ° 1 Scheduled Tribe (St oF Other (ther castes 25 well a not belonging to a nr ry) ° 1 2 caste, 55 common with several non Hind houschols) Other ost oats ° 1‘Mina Word Development 1042018) 344-357, a8 ‘ante a2 Determiants of moving aut and into being stunted in adolescence, ptstent stunting, and HAZ in vound 4 (Paslnoniows model, Mulino vg Dependent variable Moved out of bing sTuied in adolescence Moved int Being stunted in adbescnce—_ Peistently Stoned oO @ 3 ‘ender (1 027 oss O14 Irate) (os) (2, ar) Dietary dvesiy 008 “ous! 00? (10) coun) 12), Motherseducation “103, “i “0! (oon, (003), aos) West index 2008 1.6 153" 6? (085), (058) om) (031) (042) (053) Scheduled ibe 08 ose a7" (03) (ox) (053) fackward cate 022 ose war (020) (034) ost), (os) (os) os) Dbsenatons 995 995 95 Resquared - e : Wald Test Wald 72(38) = 2306 Pro > 7° 000. ‘owe: Regression include owehold size type (uranjrral) and region of esience 2s wall ae two dummies indesting Wf he community he child ved in ound 2 had ‘ees good elec ad good wate: Robust standard enor in parentheses adjusted focuser at comunity level.” p01,” p05,” pt “able na ‘Youn Lives Round 4 questions ws wo ul psychosoia outcome indices. Index ame Round survey questions used Self stem compaale with questions imam proud of my clothes: am proud fhe work have To do Teel my Gating igh forall occasions Lam proud of previous rounds of data ny shoes oof having shee Scaes1-3 ‘Agency, compatible with questions in ‘ther people in my family make al the decisions about how spend my ime: have no choice abou the ype of werk previous rounds o ta ot mus do hissort of work: ty hard can improve my tution ne; ke to rake plan or my fture stuaies Seae1-5 fd work I stay ard at schoo wl be rewarded by a Deter job i he Fate sat etesey Itsomeane opposes me. anid the means and ways to get what wane: When tam confonted with 2 problem Ian ‘ele ‘ually find several sltions I Lam in trouble, Lean ally think of 3 soltion: am confident tat Ian des eicieny wit unexpected events: can aways mane to solve ict problems ry hard enough; seas for ‘ne tose omy aims al ecomplish my goss an eminem when acing ecules bce an rey on my ‘an coping bives | cam usualy ane whatever comes my way: Thanks to my Fesurcefunes | can hale ‘foreseen tations cn seve most problems Lines he neeesary elo Peer Relations {ake tends easy am popular wt kid of my own age; Most oe kis ike me; Othe ids want me tobe their ‘als fend have moe tends than oder Kd: ave oof eds a ey Hk long with other Kids ety Pent Relations "em parents: By parents ike me: My parents and spend aft of ime together: 1 along wel with my parents My parents understand ee: Ihave cen fy a wat Io bing them up like my parent sed me My poets are easy oak or My patents and have 2 ot of fun ager eneral "am as good a most other people: Overall hae 3 to Be proud can do things as well as most ther people: ‘Seale 4 (ther people think am 2 go person: doo afmportant things, In general ike beng the way Lam: Alto things bout me are god: When {do something! dot wall The procedure adopted 0 compute the indies to st reode all elvant questions ar rece to Be postive outcomes, second, normals al sponses ores {aubtart eean anc divide by SE) and tid tle an serge ofthe reovaee-scoes aves the questions Cat have no sig ake. Qpstion follow it type sales ‘ging rom | too 5, ‘rable nt ‘Young Lives Round 3 questions use to ul ndces to acount fo being baled o teased. index Round 3 survey questions wed ‘Vetbl Buying Called you names or sworn at you Made fun of you or sme euson Pysicl Baling Punched, ike of beaten you up: Hurt you philly io any eer way (ther types af Tred to get you in rouble with ends: Made you uncomfortable by stain t you fr 2 lang ime; Rfid total to you or made other people Bulng ‘otal to you Ted to beak or damage Something of yours: Took Something without yout peemission o¢stoe fom you ‘he elevant survey questions ar Based on the sandardize Soil and Heath Assessment Pes Vcimiaton sale (ickn Schusb.Stone, © Vermisren, 2008, Chien vere asked to ihey ad experienced the 9 items buying above with esposes coded as never. Ime, 2-3 times or moc. For each child, te procedure adopted 10 Compute the indices was ost const a dummy variable that equaled | foreach of che survey questions above answers ae 2-3 ties or Aor more times Seco at les one of the question eating othe Index to Be computed eqs, then the corsponding index i aeigned (and O oer)386 Hina Wotd Development 10 (2018) 344-357 Table AS abuses check or sults ep In Tale with dotcom. ‘Dependent variable ‘Aseng Safeco sa eines Pee ans Parent relationship ‘General [tended Spcfirotion A sted contro Stunting pro reerene roup: never sunted) ‘Moved in adaescence ‘07 007 ons 025" 05 ons (0.06) (010) (008) (008) (oon (006) (08) (010) (010) (008), (0.09) (oor) Porsstenly stunted 08 07 ‘05 05 a8 ot (06) (008) (006) (oon (00s) (006) Moved out id hood oot ‘004 ‘002 ‘002 aor or Jade (0M), (006) (005) (005), (oon, (oon) Marie 035" foo 008 020" 028" 0 (006) (008) (007) (007) (00s) (005) xended Speciation & set controls Stunting pro referene group: never stunted) Moved in adlesence “8 -007 on 02a 005 004 (007) con) (07) (007) (00s) oon Moved in mid 00 007 ‘003 ‘005 03 ‘om hood (007) (010) (10) (008) (008) 009) (006) (008), (006) (oor) (008) (oor) hondadotse (oy (006), (005 (05) (oon, (005) Shoo, 2013 (oo) (00s) (005) (006) (005) 005) Recs oot ar oot) (00) Rites oon (005) pot bled “O03 (004) I potas 010 (0.05), Constant 095" aus oar on 009 “ost (020) (026) (019) (020) (028) 028) erations oo 31 930 331 sa 31 Resqured 035 oz 025 20 eat ois in aden wo reported variables, oth extended speiicat Include al cons wedi bastine specification The sins and significance ofthese unreported variables are the same 2s forthe baseline speeication Speciation B also ineludes weak 2013, two dummy vanbls conoting for household lve shocks experienced during aolescene and BMI forage 2013. Al regressions contralledfor community aed elec Robust tandard err npateathests adjusted for clustering p01, =.05, p Comparing characteristics between is who already have fspring by ae 18 ith those who dont. os ofiprinn by ae 19 No oping ‘ati (val esting for equi Ta means in (1) and @) a a Ne 1905, 1898 “185 (006) ‘Menarhe by age 12 round 2? oat ozs 415 (000) barat age 8 128 157 258 (000), Aza age 12 ar 66. “172 (008) az tage 18 ain “164 030 (042) Some stating 03s oss 967,050) Persistent stunting 10 905 079,043) Never stunted a3 250 021 (083) Wealth inde, age 8 035 oa 266 (0.00) ‘Wealth index age 12 040 oa 322 (000) ‘Wealth indoe ag 15 047 os 27 (000)" ‘Wealth index age 19, 037 050 187 (006F Coste ‘Scheduled Caste 023 ono 097.033) Backer Caste 32 026 131025) Sched Tbe on on 002 (088) Rural residence 076 053 2.40 (000 Highest edscational quien! 19 310 83,000)" * Highest qualcation:O= none. 1 = primary (as 5 2~ upper primary (dass 7 3 = matclaion erate (cass 10} = Senor Secondary Sthoo eifcte = cette, Diploma in technical edveation.‘Mina Word Development 1042018) 344-357, 357 References ‘Ao, . ¥en, A. all He Gigante. Guntupli, A.M Hort ‘cher (2013) tral bel od chido ptr Ie oul of anti 102, 319-354 ‘eran Bern, Kl, WP, Malu, J A & Watkins, 5. (2001, Son in Tongan hosel vey ata Some test for tee Aecclpingcunty samples Derr Resarth,5,79.128 -Aurna 2016 Doboyeat beer than sininda? Lngednleidence ondary versity and foo cosumpien dsprtis among hen and ase ona aman typ non 101 eh 200610007, sar vay Pete Poy, ME DL ah 5, Welle cok As Ping, & Boyden, (2512 Cena protie: Te Youn es Scud. tration oral of deg, 423), 101-708. sce h. 8 hay BUTS) lingerie Fthe dele: Howth evel idee tom ald Word Deepen 682-95. ‘eran (988) Netto health, thee nd esonaiynraouseold “ist mon cen in ore tdi orl of eae conan 28 fee ‘telah is indom,D Hadley, .Gebremaim, A Ksshun, Wl 8 Kosren "(2015 Food isc and leat sow of adnescemts J Zone malo Sc 8 Rawngs Be (11) negra teat coping coos The peal of ender iy? orl Pb ‘mmm 50) 26-200 sacle RE len, € HB, A. Caled LE, De Onis M, x, ‘Matera and Cl Undentitten Study Gebup (2008) Natal nei ‘Sscrton: Gohl se ogi expres Sn eat consequance, The Tier in) 283-350 oro, (2098) Caney, nd poverty in ni. eve f evelpment ‘omomis a 398-8 orcas tn Dats SO El TOSESSUKON Sroase', ye (20180) Yone ive: on nema Sof Cito Povey: Rnd [st etn ne Dat sere SX 6 fo 10S2598KON: Sraboen ‘ayer 2018 Youre Liven rational Stuy of Cod Poverty: Rounds 1 ST comarced Pen 202-201 [ta cleo, de. UK ata Sere SN: ahd 105258/0K08 SA 7489 9. on -oenana lS Sane, A Poy. MD, (2018), Youn ‘ion mma Sg of Chao vege Round 2013-204. (653 ‘ct Uk bats Seve S: 731 dk 1SB55uON SH 759 chang Maer, 3 fr rnthae MCCrgor 8 Powel A (202), Ey hildhod taming and str ben and col cheer ow of hd ‘yeni and chary nd od Deine #30) 775-73 Cesta NE Ars Hoe BL, Dumith S.. de les De AS (Gora) Aid lar or acelcted grown ehdhoo: A systema ‘eve: Matra and Cid Hoh four #90), 512-519. tei VG (198) Siig feltnsip in eros clral espective. Joma of ‘Marg and he or 720 Deston A rz 20) food and mtion in ni: Fat ad interpretations. rm and Fil Wendy, 42°65, Dekh & oes (1997) Per elton in adolescence: es of prenting ea ee ia ra Rien), 3. hire Bene fom the young Iver sample four develop coun Giron, ut and Eon 10) 138-19. Deron, 5.8 Since A. (2013). Halght nid dilhond and psychosocial onpeiender iste childhood: Eence om four developing coun. zon Humor Bly, 118, 26-932. Desde, (200) Does este til deine dip? A lok tie in ‘Nera nds. Ne Aron Eom Re 2) 92305, own Bs llgan Sto), ets. Vrs, (1997). re toe nora caren range? te Wea wth i 3147074199. oye Haron, Ps Heckman 8 Trem. (209) vesting ney "roman depen: Ting ad seo enc. Eznamie Haman Biba. 1) 128 spinon (2014) Young ves Round 103 Constr Fils, Young Lies “ec Note 30 Fin, Gy 8 Roce P, C. (2014), Cldhood growth, scacing, ad copitve Mcclopment acter sidence ftom the Young nes sy. The onan Joel a Nero. 10 1) 182-188. Golden Mt. (1990) complete atcvup possble fr stunted manos hen? Eopeon our of Ca! uton #8) 98-7 Geatham MeCregor 5 Cheung Cut Clee Pee. 8 SOP 8 (2007), tnterational chil development. scering_gf0up developmen {stent inthe st 5 yeu fr chien in Secopng come Toe Lace, Stas) 00-90, "2 (2008 “Reducing ation in panel sues in deeoping counties. meratinal oral of Edel, 333, (93-48, ian (2009) Posted stunting in ade choad: The case of nda "des sng ond! dats lin (2 30-38 odio J, tran}, Mali} A Mega Pe Qoourbing AR Ranier ‘Zen hz Sm A, Voss Ks Matoe (301) Ade conequenes of ron fie in ety eilod. The American Journal of incl Muito, uty. 5, ones N(2014 Yor tes an international stad of ito Poverty: Round 1, 200 data collection} sted UK Uta sevice SN 5307, 6 1.5253) UKDR-SHS07-2. Jyacandran 5 Kiziembo, 2011), Why do mothers breastfeed gl ess than ‘boys? Evdence and ptcatons for dl health nda The Quarter Jour of tonomis, 1263), 1485-1338, Jyehanaran 5. Pande, 2015, Wy ae Ion cies shat? (No. w21036) "Katona areas of Economie Research, ut, HE, Bib, N, Mile, D Santer, S.J. (1982. The eer of cone ‘hldnoadalnutiion on pubertal growth and development. The Amercan urna of Clifton 33), 527-535, ra, (2008), As Asessment of the Young Lives Sampling Approach in Andhra ‘Pradesh In, Young ives Teel Note 2. eoy J ty Ruel it, Hable} P & Hong, EA. (2015). Using hegh-forage “ieee (HAG) instead heght-foragezScores (HAZ) or the meaning ‘nese of ppt each ar wh ne es long 5 & Freese} (2014 Regression modes or cateorcal and fmt dependent "ribs using stata (34 ea). cllege Staion, Texas Sata Pres. Malu A Hodlnet J, Berman}. Martorel -uisumbig. AK 8 tla, ‘XD. (2009) The impact of inproving nutrition during eaty cided on ‘tucson among Catemalan ada, The Eamon oural, 114527) 736-7, Marsh HW, & ONL 0.5. A LL. (1984 self desrpton questions: The construct validity of mulbcimensionalslFeoncept ratings y It olesconts Journal ef Facatona Mecsrement. 2102) 153-178. Mariorell , Khan LK Schroeder, D. (1954) Reversi of stunting: ‘pidemislagcal findings In len from developing cuties. European Jura of Cnc Nation 48, 545-57. ‘oni MD. Onyangn, A W. Borghi ES. A. Nishida, & Sickman. (2007. ‘Development ef 2 WHO grown felernce for school-aged” enlden and ‘olscetefltn of the Worl Health Organon (9) £60-S57 ‘outes i Pore, (2013) caching up fom eatynutoni dees? Evidence ‘fon ral Ethiopia, Economies & Haman Boy. 112). 148-163 Pande, RP. (2003) Selective gender dferences In childhood ‘nteition and emunization irra tn: The rele sigs. Denagraphy, 4,3 395-418. ores elma, Cau A Skakeeaek NE Toppat sé Sourgigno. {2605} ‘he timing of narnal puberty andthe ages of sexual precacy Yaratins around the worl. acca tens, and changes afer mization. doce Reviews 243) 068-633, Pils K Pre. 00. and Revol, PE. 2016, xperences of Poe Bling among “Adiescnts and oid eect on Youn Adu Outeomes: Lengua Evdee ‘rom ipl, nda Peru and Vet Nan (va 3) ngacet Discussion Pape Rogol A. On Clark 1. A. a Roemich. J N- (2000) Growth abd. pst he American furl of Cal Murtian. 7202) 521-528, Rote, (1966) Generalized expectancies for inirbal versus external contol of ‘einorcemene Rycologel Monograph: Cneal and Ape 8011-28. ‘uchki, V swab Sone. M-f& Vereen, (2004) Solan Heal ASssnent (GAIA) Pychometne deveipment summa. New Have: Yale Universi. ‘Schwarze, Jerusalem, N.(2010 The general Seay scale (GSE) Ae, ‘Sess and Coping 12, 329-345, ‘pears D (2013) How Nuch Incerationa Variation in hil Height Can Sanitation ‘xplin? Wadd Bank Ply Reseaeh Working Paper Ne. 635 Sublet, B Copper. RR, Siegel P.. Sioppan . Compto,F. G8 Underwood (1804), Academie aehivement and. psyholgil adjstment in shor ciildren. The National Cooperative Grown Stady, Journal Devlprent Betavowal Pda, 15,18. steinhasen H.C, Dor H.C. & Kannenberg R (2000 The behavior protle of ‘children and ‘sdolscents"with shot satre. Journal of Developmental Behavioural Paediatrics, 2116, 23-428, Swindsle. Ay & Bilsky,P- (2000. Howshalddetary diversity score (HDDS) for easement of houehad fod acess: Ineo aude, Washington. DC: Food Sd Nutrition Techical Assistance Proje. Academy for” Educational Development ‘Taner ML (1986). Growih as a target seeking fencon. Humen Gow (op. 167-179) US: springer. Tanne, JM. (198). ets ito mon: Pascal growth frm conception to maturiy. ‘Catbridge MA Harvard Ulery Press. Up F Bes. 2, Mullan, J. Sato} (2004), Personality functioning: The fallence of satteArcives of Disease in Choa, 891) 17-21 Vicor ©. G, de Onis, My, Halla, PC, Blane, MC, Shimplon,R (2010) "Wrldwide imino grow aligning implications fo eres Preis, pp.pes 2008 Voss D. (2001), Short normal statue and psychosocial disadvantage: A crcl Teview ofthe evidence journal of Paedate Endcrtnlo and Metab, 146), Voss Bs & Sandberg. BE (2004). The pycholopcal burden of shor lature Evidence aint Ewapeon our of Ending, 151(Sppl 1), 829-533. Walkers. Chang 5, Powel c, imonof E, & Grantham MeGrege,S(2007) ary idbood stunting is assodated with poor pyehaiogial funciona ia te Molecene anes ae ee psc stration. ore Yorke, and M.-Orgando Portela (forthcoming Psychosocial sales in Yung ies ‘und or sey Section, aspatonndvahiaton,YurgLives che ne.