The height premium: a literature review and meta-analysis

Introduction

Adult height has been found to be correlated with labor-market wages across various settings^1–3. In high-income settings, a contribution to this association has been hypothesized to be due to status, prestige and self- and social-esteem which may be associated with taller stature⁴. Further in some settings, this connection has been hypothesized also to be due to physical strength and more robust health accompanying taller stature, which is favorable for manual and agrarian labor⁵. Adult height is also the culmination of nutritional exposures and growth from conception through adulthood^1,6. Linear growth faltering in childhood has many environmental and nutritional determinants that may also be associated with developmental outcomes, such as cognitive, language and motor abilities^7,8. Further, environmental conditions in childhood affect the timing of growth and the onset of the adolescent growth spurt, with children from higher socioeconomic classes achieving adult height at earlier ages. While children from less-advantaged backgrounds can experience an extended growth period, on average this does not negate adult height deficits that originate earlier in childhood^1,6,9. This evidence suggests that the relationship between adult height and wages is complex, and may be associated with and mediated by many different pathways.

There is a substantial economic literature examining the relationship between adult height and wages, also known as the “height premium” (HP), in different settings. A recent review summarized that a 1-cm increase in adult stature was associated with a 4% increase in wages for men and a 6% increase for women¹⁰. However, in this review HP estimates differ greatly among studies. Further, there are methodological difficulties in accurately quantifying the relationship between height and wages, including ascertaining whether the relationship is associative or causal, and estimating the contribution of height to wages independent of its many covariates^5,11. Different statistical methods commonly are used to examine the relationship between height and wages. Controlling for unobserved variables where possible, including through use of instrumental variables (IV) methods, in some cases substantially increases the magnitude of the association between adult height and wages compared to estimates derived using ordinary least squares (OLS) methods^12,13. The relationship between height and wages is attenuated, though usually remaining significant, when controlling for potential confounders such as family characteristics (e.g., parental education and income, number of siblings, etc.), which are precedent determinants of adult height that are also independently associated with wages^10,14. Moreover, in the economic literature it is common to adjust for mediating variables, for example, cognitive skills and schooling attainment, which also are likely to reduce the magnitude of association^10,14. However, it is important also to note that adjusting for factors that mediate the relationship between height and wages can induce mediator-outcome bias in estimates¹⁵. Further, factors like sex and country income may also contribute to differences in the effect size. As a result, it is not clear whether differences in estimates are due to statistical methods or true underlying differences among study populations.

HP estimates are an important input into economic models calculating the potential economic benefits arising from improving child height, and eventually adult height outcomes with public health and nutrition interventions. Within the past several years, such models have been developed to estimate the economic impacts of a package of childhood nutrition interventions^16,17. Recently, interest in better understanding the HP has expanded beyond the nutrition field to encompass other interventions, such as a Shigella vaccine that might mitigate infection-related growth faltering¹⁸. This attention from the enterics field to better understand and estimate the magnitude of the potential long-term productivity benefits that could result from vaccination prompted the literature review and meta-analysis presented in this paper. Our analysis describes the difference between IV and OLS estimates and examines potential effect modification of the association of adult height with wages by sex and country income category that also may contribute to differences in the magnitudes of the associations. The different estimates produced by our analyses are intended to help clarify how these differences in methodology influence the magnitude of HP estimates, as well as potential underlying differences in the magnitude of the association by sex and country income category.

Methods

Results

Study characteristics

We conceptualize the relationship between adult height, a proxy for cumulative nutritional exposures early in life, with wages in Figure 1. Table 1 presents a list of studies included in the analysis, including the HP estimates extracted from each study. Further detail on OLS and IV estimates, including a list of covariates adjusted for in each estimate and categories for IV estimates, can be found in the Supplementary Materials (Tables S1 & S2). Briefly, 9 publications presented HP estimates using the IV methodology and included data from 12 different country analyses. Thirteen publications presented HP estimates using the OLS methodology and used data from 15 different country analyses. Studies conducted using IV methods also presented 12 country analyses using OLS estimates for a total of 24 country analyses using OLS methods. Countries represented in IV studies were predominantly high-income countries (HICs) including the United States and Europe (42%) and lower-middle-income countries (LMICs) including Pakistan and China (33%), with fewer studies conducted in upper-middle-income countries (UMICs; 8%) and low-income countries (LICs; 17%). Similar to IV studies, countries represented in OLS studies were predominantly HICs (58%) and LMICs (25%), and less commonly were UMICs (8%) and LICs (8%). The age at which height and wages were measured ranged from 15 to 64 years old. Measurement of wages ranged from hourly to lifetime; more than half of IV studies (58%) and OLS studies (60%) reported hourly wages. Sample sizes ranged from 427 for an IV study in Tanzania²⁴ to nearly 450,000 for an OLS study using military records in Sweden³².

A descriptive analysis of the HP estimates is presented in the Supplementary Materials.

Figure 1. Methodological factors challenging the estimation of the contribution of adult height to adult wages.

Table 1. List of studies included in the analysis and reported height premium estimates.

					OLS estimates				IV estimates
					Low-adjusted†		High-adjusted‡
Reference	Country (year)	Data	Age height & wages measured	N	Men	Women	Men	Women	Men	Women	Wages measured
Instrumental variables studies
Behrman & Rosenzweig 2001	USA (born 1936–55)	Minnesota Twin Registry follow- up survey	39–58	808		0·65%***◊	--	0.32%*		2.17%*	Hourly
Böckerman & Vainiomälki 2013	Finland (1990–2004)	Older Finnish Twin Cohort Study	Varied	1284			3.32%***	1.41%**	12.38% (NS)	11.55%**	Lifetime
Bossavie 2021	Pakistan (2013)	Labor Skills Survey (LSS)	15–64	1419	0·78%***◊		0·46%*	--	3·56%*	--	Monthly
Elu & Price 2013a	China (2006)	China Health and Nutrition Survey: CHNS	Varied	1949			0.6%**	0.8%**	5.1%**	6.8%**	Monthly
Elu & Price 2013b	Tanzania (2004)	Tanzanian Household Worker Survey	15–60	427			1.81%***	-1.1%***	6.5%***	-4.0%***	Hourly
Gao & Smyth 2010	China (2005)	China Urban Labour Survey	16+	8919			1.00%***	0.87%***	4.50%**	7.32%**	Hourly
Heineck 2005	East Germany (2003)	German Socio- Economic Panel: GSOEP	21–50	24000			0.56%***	0.17%**	0·58% (NS)	0·34% (NS)	Gross monthly
Heineck 2005	West Germany (2003)	GSOEP	21–50	24000			0.29%***	0·07% (NS)	0.52%*	0·53% (NS)	Gross monthly
Schultz 2002	Brazil (1989)	Health and Nutrition Survey	25–54	11855			1.40%***	1.66%***	7.01%***	8·62%***	Hourly
Schultz 2002	USA (1989–93)	NLSY 1979	20–28	9257			0.45%***	0.31%**	3.62%***	6.24%***	Hourly
Schultz 2003	Ivory Coast (1985–89)	Living Standards Measurement Surveys: LSMS	Varied	2872			0.86%**	0·42% (NS)	-1·05% (NS)	-4·18%*	Hourly
Schultz 2003	Ghana (1985–90)	LSMS	Varied	6814			1.48%***	1.29%***	5.56%***	7.62%***	Hourly
OLS studies
Bargain & Zeidan 2017	Indonesia (2007)	Indonesia Family Life Survey (IFLS)	Mean=39	7878	2·54%***		1·40%***	--			Annual
Case & Paxson 2008	England (1991)	NCDS	33	9855	0·91%***	0·75%***	0·47%***	0·28% (NS)			Hourly
Case & Paxson 2008	England (2000)	British Cohort Study: BCS	30	4380	0·39%***	0·59%***	0% (NS)	0·12% (NS)			Hourly
Case & Paxson 2010	USA (1988–97)	Panel Study of Income Dynamics: PSID	25–55	31999	0·57%***	0·26%***	0·26%**	0·12% (NS)			Hourly
Case et al 2009	England (1997–2005)	British Household Panel Survey: BHPS	21–60	42666	0·71%***	0·59%***	0·16%**	0·12%*			Hourly
Heineck 2008	England (2004)	BHPS	21–50	4650	0·25%*	0·55%***	-0·05% (NS)	0·20%*			Gross hourly
Lindqvist 2012	Sweden (2006)	Longitudinal INdividual DAta for Sweden: LINDA	Varied	13446	0·56%***		0·33%***	--			Annual
Lundborg et al. 2014	Sweden (2003)	Military records	Height: 18 Wages: 28–38	448702	0·62%***		0·52%*** Ώ	--			Annual
Persico et al. 2004	USA (1996)	NLSY 1979	Height: 20–27 Wages: 31–38	1577	0·98%***		0·71%** Ώ	--			Annual
Persico et al. 2004	England (1991)	NCDS	33	1772	1·06%***		0·87%*** Ώ	--			Period varied
Rietveld et al. 2015	Germany (2002–12)	GSOEP	18–65	92248			0·51%***	0·26%***			Hourly
Sargent & Blanchflower 1994	England (1981)	NCDS	Height: 16 Wages: 23	12537			2·3%**	0·90% (NS)			Hourly
Schick & Steckel 2015	England (2000)	NCDS	33	2577	0·87%***◊	0·75%***◊	0·20% (NS)	0·00% (NS)			Hourly
Sohn 2015	Indonesia (2007)	IFLS	20–65	13243	3·58%*	4·44%*	0·75%**	1·30%**			Annual
Vogl 2014	Mexico (2002, 2005)	Mexican Family Life survey: MxFLS	25–65	3860	2·30%***		1·4%***	--			Hourly

Significance of results: 0·05%: *; 0·01%: **; 0·001%: ***

† OLS estimate adjusted with minimal controls, those included in the meta-analysis did not adjust for mediator covariates. Full list of controls for each study is presented in Supplementary Materials.

‡ OLS estimate adjusted for the most controls, including at least one mediator covariate. Full list of controls for each study is presented in Supplementary Materials.

◊ Low-adjusted estimate includes potential confounding variables and was not included in meta-analysis.

Ώ High-adjusted estimate does not include mediator covariate and was not included in meta-analysis

Estimates of the effect size

Results from the meta-analysis of the association between sex and wages are presented separately for IV studies (Figure 2) and for OLS studies that are low-adjusted (Figure 3) and high-adjusted including at least one mediator covariate (Figure 4). Results of the meta-analysis of the association between country income category and wages are presented separately for IV studies (Figure 5) and OLS studies that are low-adjusted (Figure 6) and high-adjusted including at least one mediator covariate (Figure 7).

Figure 2. Meta-analysis of the association of sex with wages, results from IV studies.

Figure 3. Meta-analysis of the association of sex with wages, results from low-adjusted OLS study estimates.

Figure 4. Meta-analysis of the association of sex with wages, results from high-adjusted OLS study estimates.

Figure 5. Meta-analysis of the association of country income category with wages, results from IV studies.

Figure 6. Meta-analysis of the association of country income category with wages, results from low-adjusted OLS studies.

Figure 7. Meta-analysis of the association of country income category with wages, results from high-adjusted OLS studies.

Overall, the pooled estimates for IV studies were the greatest in magnitude and indicated that each centimeter increase in height was associated with 3.58% greater wages (95% CI: 1.62-5.54%; I²=97.5%, p<0.001)). For OLS studies, the pooled estimates of low-adjusted OLS estimates indicated that each centimeter increase in height was associated with 1.06% greater wages (95% CI: 0.85-1.28%, I²=95.5%, p<0.001), while high-adjusted OLS estimates which included adjustment for mediators were the smallest in magnitude and indicated that each centimeter increase in height was associated with 0.57% greater wages (95% CI: 0.41-0.73%, I²=95.8%, p<0.001). Of note, there was high heterogeneity indicated by I² statistics in each analysis.

Results from the influence analysis are presented in the Supplementary Materials and indicated that several studies appeared to have meaningful influence on the overall estimates (Tables S7-S10). For the pooled estimates of the effect of sex in IV studies, after exclusion of a study from Germany²⁶ the pooled estimate for males was 4.42% (95% CI: 2.93-5.92%), while after exclusion of a study from Brazil²⁷ the pooled estimate for females was 2.58% (95% CI: -0.09-5.25%). For the pooled estimates of the effect of sex in low-adjusted OLS estimates, after exclusion of a study from Indonesia³⁵, the pooled estimate for males was 0.86% (95% CI: 0.67-1.06%), and for females was 0.52% (95% CI: 0.36-0.68%). For the pooled estimates of the effect of sex in high-adjusted OLS estimates, after exclusion of a study from Tanzania²⁴, the pooled estimate for males was 0.60% (95% CI: 0.46-0.74%), while after exclusion of a study from Brazil²⁷, the pooled estimate for females was 0.29% (95% CI: 0.05-0.53%).

For the pooled estimates of the effect of country income category in IV studies, after exclusion of a study from Germany²⁶, the pooled estimate for HICs was 2.78% (95% CI: 0.79-4.78%), while after exclusion of a study from the Ivory Coast²⁸, the pooled estimate for LMICs was 3.94% (95% CI: 1.43-6.44%). Finally after exclusion of a study from Tanzania²⁴, the pooled estimate for LICs was 6.49% (95% CI: 5.97-7.02%). For the pooled estimates of the effect of country income category in low-adjusted OLS estimates, after exclusion of a study from Indonesia³⁵, the pooled estimate for LMICs was 3.05% (95% CI: 2.03-4.07%). For the pooled estimates of the effect of country income category in high-adjusted OLS estimates, after exclusion of a study from Tanzania²⁴, the pooled estimate for LICs was 1.67% (95% CI: 1.37-1.97%).

Discussion

We conducted a meta-analysis of economic literature presenting estimates of the height premium (HP), or the association between adult height and labor-market wages, including 9 studies using IV methods and 13 studies using OLS methods with analyses conducted in primarily high-income (HIC) and lower-middle income countries (LMIC). We analyzed estimates separately based on methodology (IV and OLS) and to what degree OLS estimates adjusted for mediator variables. Our meta-analysis of IV studies found that overall, each centimeter increase in adult height was associated with 3.58% greater wages. Our meta-analyses of OLS studies found each centimeter of adult height to be associated with 1.06% greater wages in estimates which were “low-adjusted” for confounding variables (age, race/ethnicity and province/region of residence) and 0.57% greater wages in estimates which were “high-adjusted” for at least one mediator variable. As a result, this suggests that statistical methods contribute to differences in HP estimates across studies. In addition, we also found within analyses using similar methods there were differences in the magnitude of the HP association by sex and country income group which suggests there also may be true underlying differences in the magnitude of the association by study population and context. Therefore, our findings suggest that both statistical methods and underlying differences between study populations and context contribute to heterogeneity in HP estimates.

In analyzing OLS and IV results separately, we found appreciable differences in effect sizes between the two methods. Further, the pooled estimate from OLS analyses which were low-adjusted for confounding variables was nearly twice as large, on average, as the pooled estimate from high-adjusted OLS analyses which adjusted for at least one mediator variable. As a result, the pooled estimate from IV studies was over 3 times larger than the pooled low-adjusted OLS estimate and over 6 times larger than the pooled high-adjusted estimate from OLS studies. This wide range of HP estimates challenges their interpretation and has implications for identifying a narrower range of plausible estimates for use in economic models estimating the returns to investments that could improve child, and eventually adult height.

OLS parameter estimates are biased if there are confounders that should be in the model but are not and that are associated with both nutritional exposures and adult wages²⁷. It is very possible that in OLS estimates there is not complete control for confounding and there are other sociodemographic characteristics and other factors that should be in the model but that are not captured through adjustment for parental education and income, for example. Further, there is the possibility of residual confounding if the confounders are mismeasured or are included as covariates that do not fully capture the relationship. It is likely the not-included and residual confounding would lead to overestimates of the association given that poor sociodemographic and other poverty-related factors, which may not be not fully adjusted for, are more common in individuals experiencing growth deficits. OLS studies in the economic literature also often adjust for mediating variables that contribute to wages, such as cognitive skills, schooling attainment and several others. Generally, adjustment for mediators will attenuate estimates, but it can also produce mediator-outcome confounding which can bias estimates in any direction¹⁵. In the studies included in the meta-analysis, most studies adjusted for mediators like cognition and schooling that are likely the major pathways for the association of adult height with wages, and which likely outweigh any mediator-outcome confounding. Therefore, the magnitude of our pooled estimate is likely attenuated as compared to the true total effect.

The IV method corrects for unobserved confounding and attenuation bias by assuming that height is endogenous and measured with error and “instrumenting” it with other variables that are linked to one of the sources of variation in height, not related to unmeasured confounders, and only linked to the wage outcome through the effects on height (e.g., regional food prices)⁴⁰. This isolates the effect on wages of the different components of height^28,41. Identifying valid IVs can be challenging, however, and some of the assumptions made cannot be formally verified but rely on analysts’ understanding, along with intuition⁴². Further, it is important to note that IV estimates provide “local average treatment effects” that represent the effect for a subset of the population for which the instrument has effects⁴³, thereby putting into question their use to represent general population-level effects. There is controversy around use of IVs, with some critics considering the larger associations sometimes found with IVs relative to OLS methods to be “implausible”^19,35. Some analysts claim that under most circumstances both IV and OLS methods produce estimates with similar errors⁴⁴. One seminal study presented a recommended list of valid IVs, representing variables which are correlated with height but not with adult wages except through height²⁷. These included (1) variables representing a “local price of health” such as environmental or climactic conditions, and a community’s supply of health-related services and infrastructure, (2) “endowments and lifetime income constraints” including family labor market decisions and human capital investments in their children, (3) variables such as the local supply of education services which are correlated with children’s schooling but not their abilities, and (4) race and ethnicity²⁷. Compared with this list of recommended IVs, more than one-half of the IVs included in the present review were computations, or statistical calculations used as IVs, and combinations of different variables. This indicates that there may not be common agreement in the literature on which IVs to use, or potentially that choice of valid IVs differs by context, which could be a potential source of bias in IV estimates.

Our meta-analysis showed that for high-adjusted OLS estimates, but not for low-adjusted OLS estimates or IV estimates, there was some evidence that the strength of the association between height and wages differed by sex, though the magnitude of the difference was relatively small (0.34% greater for males). The significant difference in estimates adjusted for mediators of cognitive skills and schooling suggests the potential of a differential sex effect of these mediating variables on wages. This could be due to differential discrimination for shorter women, or differential schooling attainment by sex as has been suggested in previous studies^22,33,45,46.

Further, our analysis produced evidence that the strength of the association between height and wages differed by country income category for both OLS and IV estimates, though our analysis included limited evidence from UMICs and LICs. This could be due to differential quality of schooling and potential future job market opportunities among different country income strata which could influence wages and potentially could shape parents’ resource allocation decisions, particularly regarding schooling. Differences in HP among country income strata have been noted in previous studies⁴⁷ which cautioned that the result could be due to heterogeneity in measurement of wages and different methodologies among studies from different country income categories¹⁰.

This work has several limitations. This review was based on a prior review published in 2017, which may not have been conducted using fully systematic methods, including assessment of study quality and a sufficiently rigorous literature search to identify a comprehensive list of relevant studies⁴⁸. The present review would have missed any studies not included in the prior review and, as it was not systematically updated to capture any studies published in the interim, is not an exhaustive review of the literature. Future research could apply systematic review methods to the HP literature, to assess whether results change appreciably. However, when comparing our results with those of other reviews, we find a general trend of HP estimates and do not conclude that identifying more studies would be likely to alter substantially our general findings. While we were able to compare our pooled estimates of OLS and IV studies separately and discuss these general differences, future research could test these differences formally. Finally, significant heterogeneity in the methodology of included studies limited our ability to extrapolate from the meta-analyses. Finally, most of the studies that used OLS were from HICs; more studies should be conducted in low- and middle-income countries to improve the evidence base in these settings.

Despite these limitations, this review has added to the literature by applying meta-analytical methods to the HP literature for the first time, to our knowledge, to assess statistically whether the magnitude of estimates of the HP differ by sex and country income category. Our findings suggest that differences in HP estimates are attributable to both differences in statistical methods and potential differences by sex and country income. Separating HP estimates by IV and OLS methodology is advisable as the method of analysis appears to be central to the effect sizes. It is also important for researchers to understand the assumptions and potential for bias based on the selection of covariates, and it is important to recognize that adjustment for mediators will likely attenuate associations. Better understanding of the effects of height on wages can contribute to more refined economic models estimating improvements in future economic productivity due to improving growth from conception through adulthood. Measurement of these benefits also has implications for interventions addressing any of the many interconnected causes of linear growth faltering. Additional research is needed to define the association of adult height with wages taking into account potential underlying differences by context and to better understand differences in statistical methodologies to address this complex relationship.