Schooling. cognitive ability or emotional well being
Upcoming SlideShare
Loading in...5

Like this? Share it with your network


Schooling. cognitive ability or emotional well being






Total Views
Views on SlideShare
Embed Views



1 Embed 1 1



Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
Post Comment
Edit your comment

Schooling. cognitive ability or emotional well being Document Transcript

  • 1. Schooling. Cognitive ability or emotionalwell being: what drives the individual’sperception of health outcomes?Dra. Graciela Teruel Belismelis 1
  • 2. 2
  • 3. Schooling. Cognitive ability or emotionalwell being: what drives the individual’sperception of health outcomes?Dra. Graciela Teruel Belismelis UNIVERSIDAD IBEROAMERICANA 3
  • 4. 1ª edición, 2004Clasificación JEL: 112, 130D.R.  Universidad Iberoamericana, A.C. Prol. Paseo de la Reforma 880 Col. Lomas de Santa Fe 01210 México, D.F.Impreso y hecho en México Printed and made in MéxicoLa serie documentos de investigación tiene como propósitodifundir el trabajo realizado por el personal académicoasociado o adscrito al Instituto, con el fin de explorarconocimiento útil para el diseño de políticas públicas y latoma de decisiones en organizaciones sociales.Comentarios a esta serie son bienvenidos.Para más información sobre esta serie, comunicarse a lasiguiente dirección electrónica: 4
  • 5. Schooling. Cognitive ability or emotional well being: what drives the individual’s perception of health outcomes?*Graciela Teruel (UIA), Luis Rubalcava (C I D E) y Paulina Oliva (C I D E)INTRODUCTION will cause researchers to overestimate or underestimate true interactions between healthNowadays household surveys collect a broad array and other social indicators.of social indicators that measure the well being of This is the case of self reported measures ofthe population beyond traditional outcomes such as health, where people’s answer depends on theirincome or years of education. The more complex perception and understanding about their health 2and expensive fieldwork activities to accomplish status. For example, Schultz and Tansel (1997)this goal are justified by a common agreement show that more educated adults are more likelythat the well being of the population is more to report themselves being ill. Notwithstanding,than a simple characterization of economic finding a negative correlation between ill-healthoutcomes. For example, there is a large literature and education is not, by itself, evidence of thethat has shown a strong positive relationship lack of systemic bias. The true relationship maybetween levels of household income and self be even more negative if people with morereported health status. [Smith and Kington education and higher opportunity cost of time(1997), Case, et al. (2002)]. are less likely to attend a physician consultation. Nevertheless, uncovering the meaning of The most difficult problem when workingcomplex dimensionality of health measures and with health self assessment information is totheir relation with a considerable heterogeneity unravel the sign and magnitude of the systemicof social and demographic indicators is not bias. Health indicators are multidimensional andwithout its pitfalls. Not only, there are problems consequently, people’s understanding andin disentangling causal effects –directionality– knowledge about their health status involvessince most of the time there is a two-way complex and uncovered mechanisms. Forinteraction between health and economic example, people’s level of education may be a 1outcomes, but also because gathering good fundamental factor when one is to rely on selfquality information of social indicators at the assessed measures about sophisticated chronicpopulation level is not a straightforward task. disease information, than if one’s goal is to[T.N. Srinivasan, (1994)]. Collecting health data characterize people’s unobserved health statusis not the exception. On one hand, measurement through self reported day-to-day basis activitieserror in health outcomes that is uncorrelated such as ADLs. Good quality chronic diseasewith other social indicators may cause social information –when reported– may heavily relyscientists to dismiss important associations. on people’s likelihood of having been diagnosedOn the other hand, if measurement error in and their ability to have understood the diagnosishealth outcomes is not random, systemic biases [Baker M., et. al., (2001)]. But even this may turn out not to always be the case if specific-------------------- indicators of symptoms or indicators of severe* Paper prepared for the population association ofmeetings. Session: New Measurement Methods in Studiesof Health and aging. Minneapolis, 2003 --------------------1 2 See Strauss J., Thomas D., (1995) for discussion. See Kroeger (1985) for discussion. 5
  • 6. illness stem people’s awareness about their health individual’s (unobserved) cognitive ability incondition [Hill and Mamdani (1989)]. understanding her/his true health condition. The probability of phasing systemic error We close this paper by providing new evidencemay also depend on whether the inquired true on whether conditional on a person’shealth status corresponds to a stable or to a non- structural parameters, the current emotionalstable health condition. Time less-variant health well being of an individual further alters herindicators may ease people’s ability to understand capacity to provide a proper answer about hertheir own conditions –i.e., height levels in health condition. To do so, we useadulthood–. In contrast, time-variant health information on an array of mental healthconditions, such as a person’s body weight may questions and information about people’s pastprevent people’s ability to provide a proper and future health expectations as systemic biasanswer. Moreover, the complex mechanism that predictors.makes people perception deviate from the true The above analysis is based on informationhealth parameter, may become even more about a person’s objective and self reportedcomplicated if for example, life-average stable height and weight measures. Eachhealth outcomes, are for certain population anthropometric output provides a differentsubgroups not so time-invariant. This is the case health dimension. Where as height in adulthoodof a person’s height when the measurement error corresponds to a relatively stable health outcome,in the self assessed health indicator depends on a person’s weight does vary in the short run. Wethe age cohort of the population [Strauss, et. al. exploit these two opposite characteristics to(1996)]. investigate whether predetermined social We have highlighted some of the difficulties indicators are better predictors about a person’sassociated with analyzing reported health knowledge when it comes to report long termassessments. Where to draw the line in relying on health conditions, or whether an individual’sself reports is a matter of empirical analysis that short term emotional condition matters, whenis worth investigating. Unraveling the complex associated with the person’s capability of offeringmechanisms that make people’s health a true answer if the health outcome is a timeperception deviate from the truth will help us variant condition.better understand how to interpret large Section two describes the data used for thisassociations between measures of socioeconomic analysis and section three discusses our findings.status and health outcomes. This is especially Conclusions are found at the end of the paper.important since health self assessment data inmany socioeconomic and demographichousehold surveys is the only source of health DATAinformation available. This paper investigates the relation of Our results rely on unique information gathereddifferent social indicators with a person’s in the Mexican Family Life Survey (MxFLS).misconception about her true health status MxFLS is a multipurpose household survey thatwhen providing a self reported answer in a collects information about many dimensions ofhousehold survey context. We start our the well being of the Mexican population. Specialanalysis by looking at the sample selection fieldwork protocols provided MxFLS baseline withmechanism that characterizes people giving an information about the health status, socioeconomicanswer about their health condition. We turn and demographic condition of every member livingnext to unravel the size of the bias and in the household via personalized interviews. Thepossible mechanisms that exacerbate its baseline is a representative household survey of themagnitude from an statistical inference point Mexican population in both urban and ruralof view when people provide a self assessment regions. Its fieldwork activities concluded in 2002.answer about their health condition. We start To our interest, MxFLS collects height andanalyzing traditional associations and study weight information for every member living inthe correlation of a person’s age, gender and the household. In parallel, individual selfyears of schooling with the magnitude of assessment on the same anthropometric measuresproviding a true answer about her health is gathered for household members above 15status. We then depart from the existing years of age and prior to the objectiveliterature and explore the possibility of an measurement. We use information on 20 mental 6
  • 7. health questions applied to every adult in the RESULTShousehold to characterize the individual’semotional well being over the past four weeks, EXTENSIVE MARGIN ANALYSIS 3[Calderón, (1997)]. To further characterize theperson’s “emotional” perception about her health Table 4 shows the selection mechanism of peoplecondition we complement our analysis with coming with an answer about their weight andinformation on the individual’s perception about height when they are asked to provide theirher current health status relative to his/her perception. Uncovering the mechanics of thiscondition a year ago from the interview, and self selection is not trivial if the decision ofwith his/her expectations about the evolution of people in household surveys to reveal their healthhis/her health in the next 12 months. condition depends on social indicators that are MxFLS baseline provides usual correlated with health outcomes. For example, ifdemographic information about years of higher educated people are more likely toschooling, age and gender of every household participate in the interview, and years ofmember. To our advantage, the survey also offers schooling enables people to have access to betterindividual cognitive assessment information. health conditions, then relying only on selfMxFLS interviewers asked every child and adult assessment information will cause social scientistsin the household to solve a cognitive test, based not only to overestimate the population’s generalon Raven progressive matrices that require no health level –despite a survey’s random sampleliteracy [Raven et al., (1993)]. Controlling for design—but paradoxically to underestimate theyears of schooling we use the cognitive scale individual’s capacity to improve her health statusachieved by each individual (0 to 12 maximum in terms of human capital accumulation. This ispoints) to uncover how much of the the case if the true association between healthmisperception of the person’s height and weight and education is characterized by decreasing 4is explained by her unobserved ability to returns to schooling. In our sample, 42 and 34understand her true health condition, from how percent of the interviewed people agreed tomuch can be attributable to her level of human provide their weight and height perception,capital. respectively (see Table 1). Tables 1, 2 and 3 provide preliminary Results based on linear probabilitysummary statistics of the data. The regressions [columns (1) through (6) of table 4]information for this analysis corresponds to 33 indicate that males are more willing to disclosepercent of MxFLS baseline. MxFLS process of their perception about their weight and heightsystematization is scheduled to end in June, despite their age and years of education. On2003, by which results in this paper will be average females are 2.5 percent less likely toup-dated. accept knowing their weight during the interview. This contrasts with a high 11.4 percent of being less likely to provide their height measure. While the answer “I don’t know” in the survey’s questionnaire is an option, it does not reveal if it reflects a true “don’t know” or an-------------------- (indirect) refusal answer to a possible3 The mental health questionnaire is based on 20 uncomfortable question. Big differences betweenquestions that characterize the most frequent depression weight and height in gender report, suggestssymptoms for the Mexican socio and cultural females allocate more resources to investigatingenvironment. The quantification of the symptoms can about their weight than they do with respect tobe affirmative or negative, and in any case the their height.respondent can choose from out of three levels ofintensity (little, regular, and high). The instrument The age of a person seems also to play a rolescaling permits to characterize a person into four broad in determining the participation to reveal herdepression levels (normal, state of anxiety, median weight and height. Older individuals are moredepressive, and highly depressive). The mental healthinstrument has been validated in the open Mexican -------------------- 4population. Its short length, comprehensiveness and An overestimation of the true relationship betweensimplicity permits the questionnaire to be applied in health and education may arise if there are increasingfieldwork activities by non-specialized psychiatric returns to schooling: the more educated a person is, thepersonnel [Calderón, 1997]. better shape she is in to improve health condition. 7
  • 8. inclined to provide information about their bad. Since there are very few extremes, and inanthropometric measures than young adults. On order to gain precision in the analysis weaverage, one year of age increases the likelihood aggregated the two first categories into one (goodof reporting in half percent. However, while the and very good). In the regressions, “equal” is theage effect is significant, its role in defining the omitted category.self selection mechanism is much less important In table 4, columns (2) to (3) and (4) to (5)than the gender factor. It is not only small in for weight and height respectively show that outmagnitude but it also does not present a of the three variables, which measure thedifferential effect in terms of weight and height individual’s belief about her future health is what 5participation. matters. Our results suggests that holding In contrast, columns (1) and (4) suggest that schooling and cognitive ability constant, if aeducation is a very important factor determining person is optimistic about her future health, shepeople’s knowledge of weight and height. is on average 3.3 percent more willing to discloseRelative to non-literate people, individuals with her height and weight, than if she believed her1 to 6 years of schooling are 10 times more health status was not going to change. Ourlikely to reveal their weight and height. results reveal that a person’s emotional status atMoreover, for individuals with more than the time of the interview is an additionalelementary school, the probability of reporting important factor in determining her choice ofany anthropometric measure increases to 15 self selection when disclosing her healthpercent. perception, overcoming by 40 percent the gender Our results suggest that –among our social factor.indicators– education is the most important Finally, it is worthwhile pointing out thatfactor for making people to reveal their the cognitive ability coefficient is robust to theperception about their body measures inclusion of the emotional well being variables.independently on whether they correspond to a This suggests that the Raven scoring is in factshort run or to a time-invariant health condition. capturing a cognitive ability and is notHowever, the positive and significant coefficient contaminated by the individual’s capacity toof the Raven test score, suggests that not only concentrate when solving the cognitive test dueknowledge plays a role in defining the to personal stress factors.participation decision, but also the individual’scognitive ability to understand his/her health INTENSIVE MARGIN ANALYSISstatus reveals itself as an important factor forgathering self reported measures of health in a We next restrict our analysis to those people whosurvey context. felt confident about knowing their height and Conditional on demographics and cognitive body weight at the time of the interview, andability, we next ask the question of whether the study how important years of schooling,individual’s current emotional well being affects individual’s cognitive ability and her emotionalhis/her decision to disclose his/her height and well being are in determining the accuracy tobody weight. We use three variables to measure estimate her health condition.this effect. a) The individual’s scoring on the Conditional on cognitive ability, the effect ofmental health module of the questionnaire, years of schooling should tell us how, in general,previously standardized for the Mexican context knowledge affects the individual’s perceptionto diagnose the individual’s degree of depression; about his health. Conditional on education, theand two categorical variables that tell b) how cognitive ability score, should give us andoes the individual compare his health to one additional hint about how important a person’syear’s ago, and c) how does she expect her health unobserved cognitive ability is in handlingto be in the next twelve months as compared to information when it comes to calculating histoday’s. The two categorical variables allow for health condition. In parallel, our measures offive mutually excluding levels of health emotional well being will give us additional cluesexpectations: very good, good, equal, bad, and very as to whether stress factors make people’s health-------------------- perception deviate from the truth, over and5 In order to test the possibility that non-linearities could above their demographics and appreciation ofdrive our age results downwards, we tested differentregression models where age entered non linearly. The confidence about knowing their healthchange in the specification did not modify our results. condition. 8
  • 9. Analyzing the size of possible systemic attributable just to a better general knowledgebiases associated with these factors is effect is quiet impressive since we are measuringimportant for at least two reasons. First, it population averages of ordinary people whowill contribute to our understanding about described themselves literate about their bodyhow to interpret the association of health weight, during the interview.assessment measures with traditional social The negative and very significant effect ofindicators –such as schooling– when the cognitive test scoring, suggests that bettercorrelated unobservables are difficult to information is not everything despite itscontrol for. Second, it will tell us how relevance. Our results indicate that a person’sseriously we should consider the presence of cognitive ability to understand the informationan individual’s emotional condition when at hand is also important when it comes todealing with self assessment information to calculating his weight. The individual’s cognitiveestimate health outcomes. Understanding the ability in reducing the measurement erroreffect of individual unobservables when corresponds on average to a 7 percent magnitudeacquiring health data from self reported of the schooling effect.measures is of singular importance since these The negative sign of the mental healthfactors are seldom available to social scientists scoring coefficient [table 5: column (2)] suggestsin regular surveys. that people with more stress are more aware Table 2 displays the distribution of the about their weight condition than individuals of“objective” and self assessment measures. During the same age, gender, schooling and cognitivethe survey’s interview, health workers asked ability, but with less levels of anxiety. Thehousehold members 15 years of age and above to coefficient, however, accounts for only 25 gramsprovide their height and weight. This was done (11.3 pounds) of better precision and is onlyprior to taking their anthropometric measures. significant at a type I error of 10 percent.Body weight was recorded in grams and height Whether it truly shows a mild associationin centimeters. On average, individuals in our between an individual’s stress condition and hissample tend to slightly overestimate the measure ability to correctly perceive his health status oncetaken by the health worker, displaying more we control for observed and unobservedaccuracy in their perception when it comes to characteristics, or whether we are trapped byreport body weight than when describing their preliminary population observations where theheight, [66,348 g vs. 66,242 g; and 160.87 cm vast majority of the individuals in our data, scorevs. 157.52 cm, respectively]. Nonetheless, both mental health levels of 28.2 that correspond to a 6height and weight outcomes, when told by the normal situation, is still yet to be resolved. [Seeindividual, display a significant more disperse table 3].distribution than when measured by the trained To learn more about the interrelationanthropometrist, [12,219 g vs.11,052 g; and between a person’s ability to accurately perceive9.52 cm vs. 8.19 cm, respectively]. his health condition and some stress related Table 5 shows the results of systemic bias on factors, we introduce to the analysis thebody weight and height self reports at the emotional perception variables we discussedintensive margin. The dependent variable is above: a) how does the individual compare hisdefined as the absolute value of the difference health to a year’s ago; and b), how does he expectbetween self reported and measured healthoutcome. -------------------- 6 Clinical experience suggests that score values in the Columns (1) through (3) suggest that range of 20 through 35 correspond to a normal person;schooling is the most important factor, among values in the range of 36 through 45 may capture someour variables, in allowing the individual to degree of anxiety; 46 through 65 suggest medium levelsdeclare a good estimate about their weight. The of depression; and individuals scoring levels of 66returns are increasing: whereas being just above through 80 can be characterized as severe. Byilliteracy allows the individual to reduce the construction, the test allows a minimum score of 20 andmeasurement error by approximately 1.2 kg, a a maximum of 80 points. There are 20 questions whichperson who has completed more than elementary option values are 1 for a “no” perceived symptom answer, 2 for “a little”, 3 for “regular”, and 4 for “veryschool is further able to provide a self report with much or very frequently.” See Calderón (1997), for anless error of the order of 0.3 kg. We believe that explanation about the test design and its validity for thea gain in accuracy of the order of 1.5 kg Mexican population. 9
  • 10. his health condition to be in the next twelve and robust to any model specification). Thismonths as compared to today’s. Table 5, column result is in line with height being relatively stable(3) shows the results for body weight. Stress in adults, thus being affected only by one’s ownassociated with the individual’s perception about ability to report, and insensitive to regularhis/her health dramatically worsening over the measures that capture the environment andpast 12 months, has an impressive positive effect information at hand.on the person’s ability to accurately reporthis/her body weight. Relative to a person whoperceives no change in his health condition, an CONCLUSIONSindividual –who is in a worsening situation– isable to reduce the measurement error in his There is a large literature that has shown a strongperception by an astonishing range of 2.4 kg, relationship between self reported health(5.3 pounds). The large effect suggests the outcomes and traditional social indicators suchpossible presence of omitted variable bias if the as schooling or an individual’s emotional wellindividual’s perception about the worsening of being. Nonetheless, uncovering the meaning ofhis health condition is spuriously correlated with these true associations is not alwaysgreater awareness stemming perhaps from more straightforward in the presence of systemicexperience with health care providers during the measurement error related to the people’spast 12 months. While we cannot entirely rule perception about their health condition.out this possibility, it is at all odds that the This paper provides new evidence aboutspurious correlation only operates for individuals the complex, but seldom known, mechanismwhose health has worsen and not for those whose that makes an individual deviate from her truehealth has improved over the past 12 months, health condition when providing a self report.precisely because they have visited a health In particular, we analyze how an individual’sprovider. The positive but not significant general knowledge affects her perceptioncoefficient that corresponds to the answer “much about her true health, and how much thisbetter than a year’s ago”, supports this awareness is attributable to the presence ofhypothesis. In addition, the fact the coefficient unobservables –such as her cognitive ability toon stress scoring becomes not significant once we processing information, and hidden stresscontrol for the individual’s perception about his factors. We provide evidence for twooverall health condition, suggests more an important health measures: body weight andinterpretation of a positive association between height; and investigate differences in people’sstress factors and health perceptions [Farmer et accuracy to report depending on healthal., (1997)], and less a problem of spurious indicators that can be classified as long vs.correlation. short term. Although we rely on preliminary Column (3) also shows that it is the evidence, our results sink one more nail inindividual perception about his health worsening learning how to collect reliable data on socialdynamics in the past 12 months, and not his indicators and in better understanding how to(positive) expectations about his health condition interpret health outcomes when relyingin the future that matters when providing a exclusively on self reports in householdmore accurate report about his weight. This surveys.result indicates that the mechanism by which anindividual decides to report about his weight isdifferent than that determining the accuracy of REFERENCEShis report. A person’s willingness to describe hisweight is associated with a positive emotional Baker M., Stabile M., and Deri C. (2001). “What dostate of mind whereas his ability to assess his true Self reported, objective, Measures of Healthweight is only positively correlated with levels of Measure?” National Bureau of Economic Research. Working Paper 8419.stress or anxiety. Calderón, G. (1997). “Un Cuestionario para Our results on height perception diverge Simplificar el Diagnóstico del Síndromeradically with those described for body weight. Depresivo.” Revista de Neuro-Psiquiatría. (60), pp.[Table 5: columns (4) through (6)]. In the case 127-135.of height only the individual’s cognitive ability Case A., Lobotsky, D. & C. Paxson (2002).seems to matter (in the sense that it is significant “Economic Status and Health in Childhood: the 10
  • 11. Origins of the Gradient.” American Economic Smith J., and Kington R., (1997). “Race, Review. (92) 5. 1308-1334. Socioeconomic Status and Health in Late Life.”Farmer Melissa M., and Kenneth F. Ferraro, “Distress Linda Martin and Beth Soldo eds., Racial and and Perceived Health: Mechanisms of Health ethnic differences in the health of older Decline.”, Journal of Health and Social Behavior, Americans. Washington, DC: National Academy 38(3), pp. 298-311. Press. Pp 106-62.Kenkel, D.S., (1991). “Health Behavior, Health Srinivasan, T.N. (1994) “Data Base for Development knowledge and Schooling.” Journal of Political Analysis: An Overview.” Journal of Development Economy (99) 2. pp. 287-305. Economics. 44(1), pp. 3-27.Kroeger, A. (1985). “Reponse errors and o ther Strauss J., and Thomas D., (1996). “Measurement problems of health interview suverys in developing and Mismeasurement of Social Indicators.” countries.” World Health Statistics Quarterly, (38) American Economic Review. (86) 2. pp 30-34. 1, pp. 15-37- Strauss J., and Thomas D., (1995). “NutrientRaven J.C., Court J.H and J. Raven (1993) “Test de demands, income and productivity: Evidence on Matrices Progresivas. Escalas Coloreada, General y their interrelationships.” J. Behrman and T.N. Avanzada.” Paidos Iberica, Eds. ISBN: Srinivasan eds., Handbook of Development 9501260607. Economics. Vol. III pp. 1983-1916.Schultz, T.P, and Tansel A. (1997). “Wage and Labor White, H. 1980. “A heteroskedasticity - consistent Supply Effects of Illness in Cote DIvoire and covariance matrix estimator and a direct test for Ghana: Instrumental Variable Estimates for Days heteroskedasticity.” Econometrica 48: 817-838. Disabled.” Journal of Development Economics, (53) 2, pp. 251-86. Table 1 Descriptive Statistics All Sample Restricted Sample Probability of self reporting (%) 42.21 Weight … (49.39) 33.94 Height … (47.35) Demographics 0.47 0.43 Male (0.49) (0.49) Age 37.56 38.61 (17.35) (16.20) Years of education 0.11 0.09 0 years (0.40) (0.29) 0.40 0.40 1 to 6 years (0.49) (0.48) 0.48 0.51 7 years or more (0.50) (0.50) 5.97 6.24 Raven (3.01) (2.98) Observations 6,661 2,010 Notes: Restricted sample corresponds to individuals who self reported their weight or height during the interview. Preliminary statistics based on 33% of MxFLS’ sample. 11
  • 12. Table 2 Health Measures BODY WEIGHT (in g) Measured Smallest 36,800 Mean 66,292 25th percentile 58,000 Standard deviation 11,082 Median 66,400 75th percentile 74,800 Largest 87,800 Self Reported Smallest 30,000 Mean 66,348 25th percentile 58,000 Standard deviation 12,219 Median 66,000 75th percentile 65,000 Largest 184,000 BODY HEIGHT (in cm) Measured Smallest 130.3 Mean 157.58 25th percentile 152 Standard deviation 8.19 Median 157.45 75th percentile 163.8 Largest 173.5 Self Reported Smallest 100 Mean 160.87 25th percentile 155 Standard deviation 9.52 Median 161 75th percentile 168 Largest 199 Preliminary statistics based on 33% of MxFLS’ sample.12
  • 13. Table 3 Mental Health Scoring All Sample Restricted Sample Health Perception & Health Expectations How is your health compared to the way it was one year’s ago? 0.258 0.258 Very good/Good (0.438) (0.437) 0.625 0.622 Equal (0.484) (0.485) 0.112 0.117 Bad (0.316) (0.321) 0.003 0.003 Very bad (0.060) (0.055) How do you expect your health will be compared with today’s? 0.376 0.413 Very good/Good (0.484) (0.492) 0.559 0.524 Equal (0.496) (0.499) 0.061 0.061 Bad (0.241) (0.239) 0.002 0.002 Very bad (0.486) (0.043) Mental Health scoring (scale 20 to 80 points)* Smallest 20 Mean 28.20 25th percentile 23 Standard deviation 7.28 Median 26 75th percentile 32 Largest 80Notes: * See footnotes number 3 and 6 for Mental Health scale interpretation. Preliminary statistics based on 33% ofMxFLS’ sample. 13
  • 14. Table 4 Extensive Margin Analysis Probability of self reporting weight or height WEIGHT HEIGHT (1) (2) (3) (4) (5) (6) 2.3917 2.595 2.5559 11.3408 11.0614 11.0052Male [1.1579]** [1.2040]** [1.2040]** [1.0908]** [1.1318]** [1.1313]** 0.6498 0.6643 0.6491 1.0908 0.5335 0.5242Age [0.0483]** [0.0487]** [0.0498]** [0.5265]** [0.0454]** [0.0463]**Years of education 10.0531 10.2012 10.1056 11.6202 11.5098 11.53241 to 6 years [2.0967]** [2.1176]** [2.1194]** [1.6511]** [1.6697]** [1.6694]** 14.6103 15.0007 14.7461 23.6144 23.297 23.12267 and more [2.3960]** [2.4202]** [2.4276]** [2.0413]** [2.0656]** [2.0683]** 1.2946 1.2789 1.2734 1.9414 1.9387 1.9348Cognitive test score [0.2231]** [0.2258]** [0.2258]** [0.2094]** [0.2124]** [0.2126]** -0.0379 -0.0513 -0.1436 -0.1434Mental health score … … [0.0834] [0.0880] [0.0743]* [0.0786]*How is your health compared to one year’s ago? -0.2275 … 0.1454Very good/Good … [1.3558] … … [1.2631] 0.5809 -1.04Bad [2.1072] … … [1.9027] -1.2667 -1.5725Very Bad … … [10.9357] … … [9.3255]How do you expect your health will be in the next twelve months as compared to today’s? 3.2995 3.5801Very good/Good … … … … [1.2650]** [1.1816]** 2.4183 3.5014Bad … … [2.7149] … … [2.4117] -1.0077 -10.6762Very bad … … [13.4549] … … [9.2429]Observations 6,808 6,661 6,661 6,812 6,664 6,664R-squared 0.1315 0.1316 0.1325 0.1936 0.192 0.1935 42.44 36.56 18.78 96.84 80.34 41.57F (all covariates) [0.0000] [0.0000] [0.0000] [0.0000] [0.0000] [0.0000]Notes: Linear probability regressions. Dependent variables multiplied by 100. All models include community fixedeffects to control for locality health and school related infrastructure. Robust to household clustering andheteroscedasticity standard errors. Standard errors in brackets below coefficients, p-values below hypothesis tests. **Significant at • 0.05; * significant at • 0.10. 14
  • 15. Table 5 Intensive Margin Analysis Dependent variable: Absolute Difference between self reported and measured weight and height WEIGHT HEIGHT (1) (2) (3) (4) (5) (6) 540.886 474.3184 480.2646 -0.4445 -0.4523 -0.4198Male [194.7635]** [199.3113]** [199.0631]** [0.2616]* [0.2757] [0.2759] -1.9984 -2.4376 0.0222 0.0068 0.0067 0.0075Age [8.2520] [8.3995] [8.5076] [0.0101] [0.0103] [0.0107]Years of education -1,231.46 -1319.56 -1309.47 -0.3882 -0.3088 -0.20071 to 6 years [648.2279]* [636.6658]** [632.4851]** [0.7653] [0.7780] [-0.7407] -1424.37 -1555.99 -1558.32 -0.6122 -0.5025 -0.38157 and more [670.9362]** [660.2482]** [661.7705]** [0.7636] [0.7739] [0.7334] -97.8392 -108.5464 -106.1697 -0.1173 -0.1171 -0.1182Cognitive test score [35.9378]** [36.1874]** [35.9297]** [0.0416]** [0.0427]** [0.0432]** … -25.0688 -20.9701 … -0.0018 -0.0096Mental health score [14.9051]* [16.3522] [0.0195] [0.0209]How is your health compared to one year’s ago? … … 275.5854 … … 0.3386Very good/Good [252.9378] [0.3264] … … -297.9212 … … -0.0688Bad [333.1089] [0.6482] … … -2406.67 … … -0.4083Very Bad [786.4449]** [2.1242]How do you expect your health will be in the next twelve months as compared to today’s? … … -93.6661 … … -0.0111Very good/Good [207.6075] [0.2630] … … 40.4054 … … 1.1401Bad [547.3454] [1.0012] … … -569.98 … … 0.0001Very bad [823.4151] [0.0002]Observations 2,655 2,611 2,611 2,056 2,010 2,010R-squared 0.0422 0.0443 0.0458 0.0555 0.0577 0.0577 6.07 6.67 4.35 3.24 2.47 1.42F (all covariates) [0.0000] [0.0000] [0.0000] [0.0064] [0.0223] [0.1550]Notes: See notes Table 4. OLS regressions. Weight measured in g, height in cm. 15
  • 16. Vélez, F. y De la Torre R. (1993) "Ladesigualdad en la distribución de la tierraejidal en México", Documento de Trabajo delDepartamento Académico de Economía, DT-5,Instituto Tecnológico Autónomo de México. Wolfson, Michael C. (1994) "Wheninequalities diverge", The American EconomicReview, Papers and Proceedings. Wrigth, E. O. (1986) "What is middleabout the middle class?", in Roemer, John. E.Analytical Marxism, Cambridge UniversityPress. 16
  • 17. 17