Scarring and Selection in the Great Irish Famine

1. Introduction

One of the universal features of famines is the immediate and devastating human tragedy they bring. Another, apparently, is the subsequent deterioration in the health and human capital of survivors. A plethora of case studies already document the scale of catastrophic famine events measured using adult stature. Figure 1 depicts our review of these other anthropometric studies. We plot famine-period excess mortality against the impact on the average societal stature of the famine-born cohort. The figure reveals the smallest drops in stature occur in populations exposed to the most severe incidents of famine-induced hunger and disease. This highlights the fact that surviving populations are selected populations which have fundamentally different characteristics to those that died or moved away, something that the studies depicted in the figure do not themselves tend to dwell on very much. But it is the health and human capital of such selected residual populations that is crucial in understanding a society’s subsequent recovery. Those that perished, or migrated away, do not, by definition, contribute directly to the wealth of the nation. Our contribution is our ability to use individual-level data to investigate this selection effect.

• Download figure
• Open in new tab

Figure 1:

Change in average adult height and mortality in 14 famines

Sources: Excess mortality figures: Ó Gráda (2007); corresponding height changes, defined as the difference in final adult height between pre-famine and famine-born cohorts: Bengal 1942-44 (Guntupalli and Baten 2006); China 1877-79 (Baten et al. 2010); China 1927 (Morgan 2004); China 1959-61 (Gørgens et al. 2012); Ethiopia 1985-86 (Dercon and Porter 2014); France 1860-90 (Banerjee et al. 2010); Germany 1914-18 (Blum 2011); Greece 1941-42 (Valaoras 1970); India 1972-73 (Guntupalli 2007); the Netherlands 1944-45 (Van Wieringen 1972); Uganda 1980-81 (Umana-Aponte 2011); USSR 1921-22, USSR 1932-33 and USSR 1946-47 (Wheatcroft 2009).

We study one of the most severe and lengthy famines in modern history, The Great Irish Famine of 1845-52: one million people died, and another million emigrated (Mokyr 1985). When deaths are considered relative to overall population levels, the Great Irish Famine overshadows practically all other famines: 12 per cent of Ireland’s population perished (Ó Gráda 2007: Tab. 3). This famine also led to devastating structural change: the catastrophe meant that the island’s potato-based economy was simply no longer viable, and it intensified pre-existing patterns of migration from which the island has not yet recovered. We assess the long-run consequence of this catastrophic event on the health and wellbeing of those born in famine conditions, remained in Ireland and survived into adulthood.

Adult stature reflects a population’s living standards in early life, with the crucial period being from inception to the age of two (Currie and Vogl 2013).¹ The average adult height of the generation born in distressed conditions is a measure of a society’s health capital used widely by health and development economists – and economic historians – researching catastrophic events (Fogel 2004; Floud et al. 2011; Deaton 2013). The biological mechanisms underpinning the link between early childhood exposure to acute malnutrition and reduced adult stature relate to fetus growth in utero.² Known as the “Barker thesis”, or “fetal origins hypothesis” (FOH), scientists link low birth weight with a variety of poor health outcomes into adulthood besides short stature (Barker 1990, 2004; Calkins and Devaskar 2011; Almond and Currie 2011). Scholars use adult stature as a measurable proxy for these other health conditions.

Continuing in this anthropometric tradition, we focus on average societal stature as our measure of health and human capital in the context of the Great Irish Famine (henceforth referred to simply as “the Famine”). Our individual-level data contain over 21,000 subjects born before, during and after the Famine in locations across Ireland which varied in their exposure to famine-induced nutritional deprivation and disease. These micro-data are drawn from archival records of prisons located in Dublin, Ireland’s capital city, and Clonmel, in rural Tipperary. Indeed, one of our contributions is the collection and use of these fine-grained data; previous explorations of this, and other, famines do not enjoy such large individual-level datasets.

The Famine’s root cause was an exogenous ecological shock – Phytophthorainfestans (potato blight) – which decimated the main source of food of the majority of the population. Those born during the Famine suffered this ecological shock in a crucial developmental stage and experienced devastating and often permanent health consequences, ranging from chronic disease to early death. The catastrophic event led to structural changes in food production as the potato was rendered an unreliable source of sustenance until the end of the nineteenth century, when the (Bordeaux mixture) treatment for potato blight was first discovered and applied. Consequently, those born after this famine enjoyed a very different diet than earlier generations.

Famines exacerbate the mortality of specific elements of the population (young and old). They also cause reductions in birth rates as people are unwilling or unable to procreate. Afterwards, birth rates rebound (Ó Gráda 2009). Both the selection in mortality and fertility are issues which can affect the attributes of a population after a famine. A poor childhood nutritional and disease environment can therefore have two effects that work in opposite directions: scarring and selection (Deaton 2007; Bozzoli et al. 2009). In the scarring outcome, surviving children do not realise their adult height potential, with the degree of the reduction depending on the severity of the nutritional and disease shock. This, in turn, may lower societal average height. By contrast, selection occurs when a high disease and low nutritional environment raises the survival cut-off threshold and fewer (low-height/low-health) children survive into adulthood, potentially increasing the average health and height of survivors (Alter 2004).

Our principal contribution derives from our historical context, which is uniquely suited to the study of catastrophic events in general, and famines in particular, because the strong spatial heterogeneity in its impact permits us to conduct an ecological identification of exposure during early childhood. Our micro-study offers new insights because competing scarring and selection effects may be hidden in the aggregated national studies typical of this literature. We contribute by attempting to differentiate scarring from selection effects. Indeed, we find the same famine severity/health capital correlation depicted in Figure 1 occurring within this one famine episode. Most famine deaths were recorded in rural parts of Ireland; direct famine-related mortality in Dublin, the island’s largest urban centre, was absent. And while we find few signs of early childhood malnutrition in the surviving population born in rural Tipperary, we do find evidence of such scarring of those born in Dublin (Figure 2). Our estimates of the extent of this scarring range between 0.3 cm (Table 4, Model 1) and 2.5 cm (Table C1, Model 1).³ These estimates represent large biological and economic effects: Baten and Komlos (1998) associate a 1 cm decline with a reduction in average life expectancy of 1.8 years; McGovern et al. (2017) link a 1 cm decrease with a 4 per cent fall in wages.

• Download figure
• Open in new tab

Figure 2:

Change in average adult height in the Great Irish Famine

Notes: Error bars depict 95% confidence intervals; reference category is an adult, pre-famine-born illiterate male Catholic who committed an offence against property (non-violent) and was imprisoned before 1877; see Table 4.

Famines are not just a historical curiosity; policymakers are once again interested in understanding the impact of this catastrophic risk. Historically, absolute food availability was more important than distribution and entitlement issues in causing famines (Alfani and Ó Gráda 2018; cf. Sen 1977). Extreme climate change may mean the same becomes the case in future. We contribute by showing how a historical case can be used to think about questions of potentially contemporary relevance.

This paper proceeds as follows. Section 2 provides the necessary background on the Great Irish Famine. Section 3 sets out the issue of selection in individual-level data and how this affects studies of the FOH. Section 4 outlines our data source. Section 5 defines our empirical strategies, while Section 6 presents our main findings, and Section 7 introduces a set of additional tests. Section 8 discusses our results and restates our overall contribution. A set of appendices report on the details of a series of robustness exercises and extensions.

2. Ireland’s Famine

Ireland experienced Western Europe’s last great subsistence crisis: back-to-back harvest failures between 1845 and 1852 caused by the fungus Phytophthora infestans, or potato blight. Many farmers had come to rely exclusively on the Irish Lumper variety of potato because it was particularly well suited to the island’s poor-quality soil (Bourke 1993). The Lumper proved to be very susceptible to blight, and there existed few alternative sources of sustenance for the majority of rural folk; rural Ireland had become a monoculture. Approximately 20 per cent of Ireland’s population either perished or migrated as a result of this blight-induced famine, reducing the island’s population from 8.17 to 6.53 million between the recorded censuses of 1841 and 1851 (BPP 1851). Then, in the post-Famine period, the potato became a much less reliable source of food security due to the continued presence of blight. As a consequence, Ireland’s diet shifted to wheat-based sources of carbohydrates (Clarkson and Crawford 2001).

Mokyr and Ó Gráda (2002) estimate about half of the deaths during the Famine were associated directly with malnutrition, while the other half were caused by indirect effects on personal behaviour and social structure. There were principally five famine-related illnesses: dysentery, diarrhoea, dropsy, starvation, and “fever” - typhoid, typhus and relapsing fever (MacArthur 1956: pp. 265-68). Famine mortality figures were highest in Munster and Connacht (west and south-west of the island), followed by Leinster (east). The lowest mortality was observed in Ulster (north-east). Figure 3 locates the regional impact of the Famine in terms of excess mortality, using county-level estimates from Mokyr (1985).

• Download figure
• Open in new tab

Figure 3:

Famine maps of Ireland, with geographic location of two prisons

Notes: Excess mortality rates are an average annual estimated total; emigration rates are totals expressed relative to the 1841 population.

Sources: Mortality data from Mokyr (1985), migration data from Ó Gráda and O’Rourke (1997), and GIS shapefile adapted from Gregory and Ell (2004).

In terms of this study, the relevant populations relate to the counties in which our subjects were recorded, and earlier, born. We have individuals born across the island. But the majority hail from two locations: the Munster county of Tipperary, and the Leinster county of Dublin. Contrasting population trends can be seen. The population of Tipperary was ravaged by the Famine: in 1841 the population stood at 0.44 million, but by 1851 this had fallen by a quarter to 0.33 million. Lower and upper bound estimates of excess mortality in Tipperary during the Famine are 24 and 35 per 1,000 population (Mokyr 1985). This compares to the all-island median of between 19 and 26 deaths per 1,000 (see Table 1).

Meanwhile, Greater Dublin (city and surrounding county) had a population of 0.37 million in 1841. This increased slightly to 0.40 million by 1851 as migrants from elsewhere in Ireland fled the famine conditions. The increase in the population of Greater Dublin masks the growing “slumification” of the city. Dublin’s plight was somewhat atypical for cities in the UK in the

Nineteenth century as its population was not sustained by growing industrialisation; the city was increasing in population size without consummate improvement in the quantity or quality of its housing stock. The slumification also had an influence on the general disease environment, so much so that Dublin had one of the lowest life expectancies of comparable cities (Prunty 1998).

As the population of Greater Dublin expanded over the decade 1841-51, an older strand of literature saw Dublin (as well as Cork City and Belfast) as being ‘unaffected directly’ by the Famine (Lynch and Vaizey 1960). Ó Gráda (1997: pp. 157-193) challenges this view, but concedes the famine mechanism was indirect. One such mechanism was increases in food prices and resulting falls in entitlements; the inhabitants of Dublin were less likely to be affected by an absolute decline in food availability as the urban diet was as much based on wheat (bread) as potatoes (Ó Gráda 1999: p. 163). Data on urban wages suggest a slight decline in real terms (D’Arcy 1989; Kennedy and Solar 2007). The second mechanism was immigration of famished rural-folk into the city. Indeed, Ó Gráda (1999) finds any increased urban mortality during the peak famine years was not driven by native Dubliners; rather, it was caused by rural immigrants dying of fever and dysentery. Together, these indirect famine effects make the Dublin-born population a useful comparison group for the directly-exposed rural population of Tipperary.

Internal and international migration was one of the main non-governmental responses to famine conditions (Kenny 2000; Donnelly 2001; Ó Murchadha 2011). Migration enabled the famished to go to sources of food located in urban centres in Ireland and beyond. The Irish-born population of Britain almost doubled between 1841 and 1851, from 415,725 to 727,326 (Macraild 2006). North America was the main destination for international migration (Akenson 1996). After the Famine, Ireland had the highest migration rates of Western Europe (Hatton and Williamson 1993, 1994). As with famine-related mortality, famine-induced migration differed regionally. Ó Gráda and O’Rourke (1997) estimate Tipperary experienced one of the highest emigration rates relative to mortality, with a ratio of 1.18, whereas Dublin was an anomaly in the sense it was a net-recipient of migrants (see Figure 3).

3. Understanding Selection

In Lumey et al.’s (2011) comprehensive review of the effects of famine on prenatal and adult health outcomes, the most consistent findings across studies relate to body size (height and BMI), diabetes and schizophrenia. The reviewed studies share many similarities with our own: they are observational and not experimental in nature; they focus on outcomes at future points in time; they are cohort studies; and they use population-based registries to trace people. Lumey et al. (2011) outline the analytical strategies which have been applied to date, including simple cohort analysis, difference-in-difference estimation and sibling designs. Across studies, definitions of famine exposure are ecological rather than based on individual food consumption as such data are unavailable. Primary exposure to food shortages are difficult to disentangle from other associated famine effects within any one famine episode. Furthermore, chronologies of famines vary from the well-defined (the Dutch Hongerwinter famine lasted exactly six months) to the much-less-so (seasonal famines in The Gambia between 1949 and 1994).

A key issue in the literature on the health effects of early-life exposure to famine relates to scarring and selection, two forces which can co-occur, but to different degrees depending on the severity of the famine. In the literature thus far, the issues surrounding competing scarring and selection effects has probably been most evident in studies of the Leningrad siege. In their analysis of a sample of survivors of this 872-day German siege of St Petersburg during WWII, Stanner et al. (1997) found no difference between exposure in utero and exposure during infant life on various health outcomes. However, the possibility of selection bias emanating from selective mortality was not acknowledged (Rich-Edwards and Gillman 1997).

Probably the most studied historical famine is the Chinese Great Leap Forward famine of 1959-61, where scholars have attempted to address the problems surrounding scarring and selection using different sources and methods. Gørgens et al. (2012) look at growth stunting in this famine, using data from the China Health and Nutrition Survey. Their methodology attempts to identify scarring and selection by linking the heights of adults born before, during and after the Chinese Great Leap period in famine affected and unaffected distrincts with their own children born post-famine. The logic here is that these children only inherit the selection (genotype) and not the scarring (phenotype). They find the famine cohort was stunted by 1 to 2 cm, but that this was offset by selection of equal magnitude, leading to no overall effect on the average (Gørgens et al. 2012: Fig. 1).⁴ However, Gørgens et al. (2012, p. 103) caution that their result may be biased upwards or downwards due to the dynamics of internal migration.

Elsewhere, Kim et al. (2016) also look at the long-run health impact of the Great Leap Forward, this time using the China Health and Retirement Longitudinal Study. They find cohorts born during the famine years had poorer health and lower cognitive ability. They outline various selection issues (selective mortality, fertility and migration) which may influence their results, but as they are unable to address these empirically they instead suggest interpreting their findings as a lower bound of their true effects. Using the same dataset, Xu et al. (2016) attempt to see whether fetal famine exposure had a long-run health impact, measured by biomarkers and anthropometric characteristics. Their innovation is to use multiple estimation strategies which take account of selective mortality. Their primary method is a simple cohort difference, which compares the cohort born during the famine with cohorts born before and after. Using this cohort approach indicates a famine effect, but this effect disappears with their alternative estimators.⁵ They conclude this to be evidence of selective mortality; those most likely to have been affected by the famine had already died (“survival of the fittest”).

As a way to overcome sample selection biases from selective mortality in famines, scholars have attempted to test the FOH using non-famine natural experiments, such as pandemics and income shocks. Almond’s (2006) landmark study uses samples from the 1960, 1970 and 1980 US censuses to ascertain the impact of prenatal exposure to the influenza pandemic of 1918-19. He finds increased likelihood of disability, lower income and lower education of those exposed to the pandemic in utero (cf. Brown and Thomas 2018). Meanwhile, Banerjee et al. (2010) focus on the long-run health effects of income shocks at birth, proxied using adult male height and life expectancy. Their empirical strategy exploits the differential incidence of phylloxera, an insect which destroyed large segments of the French grape crop between 1863 and 1890. They use a difference-in-difference methodology to compare children born in phylloxera-affected areas with those born before and after phylloxera exposure. Every additional hectare of land used for vineyards is associated with a 3 per cent decline in height, approximately 1 cm (Banerjee et al. 2010: p. 723).

Our goal is to understand the effect of the Famine on the surviving Irish population. We therefore need to identify the potential influence that selection has on the question we pose. But we also need to think about the additional impact of selection on our underlying data source. We have identified the main sources of selection that affect our study in Table 2: selective mortality (S1), selective migration (S2), selection into crime (S3), and selection into prison (S4). S1 and S2 are different responses to famine: S1 is a biological phenomenon; S2 is socially determined.⁶ In terms of S1, the youngest and oldest are more vulnerable and comprise the majority of mortality in normal years in pre-epidemiological transition societies. Famine would exacerbate this pattern; we would expect the young and the old to be disproportionately affected by malnutrition and disease. To address the selective mortality, we control for famine intensity proxied by famine-period deaths. Meanwhile, S3 and S4 would bias results downwards as they imply only certain segments of society are represented in our data source.

It is possible to identify the areas of the island and the demographic groups that were most affected by the Famine in terms of mortality (S1). Evidence from archaeological studies provide indications as to the selective nature of famine mortality conditions in Ireland. The main archaeological evidence comes from the excavation of the Kilkenny Workhouse burial ground, 33 km from Clonmel (Geber 2016).⁷ A high proportion of skeletal remains exhibited evidence of scurvy rather than any infectious disease, and this disproportionately affected the young.⁸ Comparisons with other burial sites show the skeletal remains in Kilkenny had a much less varied diet (Beaumont et al. 2013).⁹

However, the representativeness of the data is uncertain due to the so-called “osteological paradox”: demographic non-stationarity, selective mortality, and hidden heterogeneity in risks (Wood et al. 1992; Beaumont et al. 2015). From the perspective of our study, the most important of these is selective mortality, as osteological data can only provide information about those who died at a given age, but nothing about survivors. For example, deaths of children are selective as those most frail are more likely to die. Those who survive early life conditions are more likely to survive longer. These issues aside, the archaeological evidence suggests famine mortality was highest amongst the young, and survivors were a highly selected group.

Migration was an adaptive (and selective) response of the population to the Famine. If the poorest and weakest migrate from the affected areas, then S2 conflates with S1 and negates the impact of S1 and mortality more generally. Figure 4 plots the mortality rates against migration rates for the Famine period. While there is no obvious relationship, it does highlight the implications for our study as Dublin has net inward migration whereas Tipperary has net outward migration.

• Download figure
• Open in new tab

Figure 4:

Famine mortality and famine migration

Sources: Authors’ calculations using methodology and sources derived from Ó Gráda and O’Rourke (1997).

Understanding the composition of migrants (e.g., were they the poorest, the richest, or of median income) helps understand the direction of the selection effect. Ó Gráda and O’Rourke (1997) document that emigration was beyond the reach of the very poorest, although this poverty trap was somewhat alleviated by remittances from first-wave migrants (cf. Schrier 1958). Another outlet was internal migration. These migration flows were directed towards urban centres. In this case, migration from the heavily affected rural areas would lead to selection (under-estimating the true effect) and migration to urban areas would lead to scarring (over-estimating the true effect).

There is a wider debate ongoing in economic history on migrant selection. Recent work by Connor (2019) using US and Irish censuses finds evidence of negative selection of Irish migrations, while Collins and Zimran (2019) find intergenerational convergence of Famine-era Irish migrants to the US in terms of occupational status. More careful comparison of anthropometric datasets either side of the Atlantic is needed to tease out the full ramification of this selection – including whether international migrants hailing from Ireland’s urban and rural areas were relatively better off than their non-migrating peers. Given both the archaeological evidence and documented migration patterns, it is evident that some element of both S1 and S2 is present, but would differ depending on urban and rural compositions. Our analysis of nationally-averaged data (composing both urban and rural samples) would possibly see the effects of S1 and S2 averaged out.

As for the prison data that we use to test the effect of the Famine, there are issues of selection into crime (S3) and then selection into prison (S4). Effectively the selection issue here relates to the representativeness of prisoners. This has recently become a key concern in the anthropometrics literature (Bodenhorn et al. 2017): were prison inmates representative of the general population, a criminal underclass, or selecting into crime differently over time? McLaughlin et al. (2021) present and discuss the various selection issues surrounding the prison data used in the present study. They find any changes in selection into crime across time were more likely to have been the result of institutional (changes in criminal classification) rather than economic factors (various business cycle indicators), but do find that there is a relationship between the two during the period of the Famine.

Our strategy to address S3 is to “sample” data from the prison population incarcerated only after the Famine concluded, thereby excluding the selected population that can distort our isolation of a famine effect. In terms of S4, it could be possible that specific crimes and the people who commit such offences are more likely to be represented in our sample. We control for the type of crime committed in our analysis to address this concern. We find a disproportionate number of those incarcerated were not doing time for serious crimes, but were rather incarcerated for short spells for the social offence of “drunkenness”. Given the nature of this crime, which only began to be enforced in the 1870s (McLaughlin et al. 2021), we argue that drunkards are likely to be broadly representative of the wider population of Ireland.

We test the impact of selection effects S3 and S4 more rigorously by comparing the observable attributes of the prison population with the corresponding census waves and re-weight our estimates in accordance with how our sample deviates from census observables. Appendices to this present paper include econometric specifications in which we attempt to further address the issues surrounding the representativeness of prisoners. These robustness exercises do not alter our conclusions.

4. Data

To gain any insight into the health conditions of the Irish population requires the use of an archival source of data in which stature was consistently recorded for cohorts born before, during and after the Famine. In terms of generating new evidence, our contribution is the use of anthropometric information extracted from one such source: prison registers. These unique and rich historical documents facilitate a direct inference of the impact of famine-related nutritional and dietary change. We sample cohorts born before, during and after the Famine from the complete population of prisoners incarcerated for any period of time between 1854 (after the Famine had ended) and 1910 in two locations in Ireland: Dublin and Clonmel (locations indicated in Figure 3).¹⁰ These registers provide detailed information on every prisoner who entered the prison system on categories including their height, age, crime, location of birth and residence, occupation, religion and level of literacy.¹¹ The data were recorded by prison administrators and are standardised in nature; new inmates were required to undergo an obligatory medical examination by a prison doctor (Breathnach 2014). Our data do not constitute only hardened criminals; “drunkenness” was criminalised and punishable with a custodial sentence.¹²

Today, stature is widely used as an anthropometric indicator of the health status of a population in various literatures (e.g., Fogel 2004; Floud et al. 2011; Deaton 2013). Heights reflect health and living standards from an outcome-oriented viewpoint. The mean stature of a population is a function of both official and unofficial income, including subsistence farming and public goods provisioning (Steckel 1995). Heights are also sensitive to income inequality (Deaton 2008). The height of a generation is determined around the time of its birth, with health standards of mothers as well as nutritional and health standards of infants playing a crucial role (Eveleth and Tanner 1976; Steckel 1995).

Deaton (2007) argues cross-country evidence on height is difficult to interpret. His case in point is Africa, where the population is the world’s poorest but also the tallest of all developing regions. He disregards genetics as an explanation in the African case, suggesting instead selection may be greater there than scarring. He also highlights the variation in heights across social groups within poor countries as evidence against a genetic explanation. Similarly, in our study we disregard genetic determinants as we are focusing on the population of one specific island which shares a common “gene pool”. An additional advantage is our sample is drawn from a single social group: the least successful in Irish society, those selecting or selected into the criminal population and subsequently caught, convicted and incarcerated as punishment for their crimes.

Over the period of our study, there were three “prison regimes” (following McLaughlin et al. 2021). The first, from 1791 to 1853, saw the removal of a segment of criminals from the UK as a punishment for what were deemed the most severe crimes: over this period, 26,500 convicts were transported to Australia (Kilcommins et al. 2004: p. 17).¹³ Transportation was replaced with penal servitude and this was administrative change occurred in 1877, when prisons throughout Ireland were centralised under the General Prisons Board. This resulted in a change in the classification of prisoners being held in Kilmainham and Clonmel: they were to be used for ‘untried and prisoners under sentence, males for sentences not exceeding 12 months and females not exceeding 6 months’ (BPP 1878-79, p. 37). From 1877 onwards, prisoners serving longer-term sentences were sent to specialist convict prisons: Lusk (Dublin), Mountjoy (Dublin), and Spike Island (Cork). McLaughlin et al. (2021) conclude there is little evidence of selection into crime across variables other than these institutional changes.

The personal characteristics used in our study include height, age, year of birth, literacy status, religion and nature of crime committed.¹⁴ Our key variable used to identify any possible famine effect is average adult height. For Kilmainham Gaol, mean male height, after excluding children and adolescents below the age of 17, is 168.1 cm (see Table 3 for summary statistics). Height in Clonmel is somewhat higher, at 170.0 cm. This reflects generally superior biological living standards in rural Ireland.¹⁵ The average year of birth is 1850, with sufficient variation to capture living standards before, during and after the Famine. At point of incarceration, males were on average aged 33.1 years in Kilmainham and 38.5 years in Clonmel. For Clonmel Gaol Protestants make up 1.7 per cent of the population, while for Kilmainham Gaol this is 9.4 per cent, reflecting the larger share of Protestants in the Greater Dublin area. The data provide for each individual their ability, separately, to read and write; we classify individuals who are able to read and write as “literate”, and “illiterate” otherwise.¹⁶ We find inmates of Kilmainham Gaol were generally more likely to be literate than their rural counterparts.

A set of control variables is based on the type of offence for which an individual was imprisoned, following the example of Blum et al. (2017). Offences are distinguished into seven sub-categories, adapting a contemporary criminal classification: offences against the person, violent and non-violent offences against property, begging and prostitution, drunkenness, rioting, treason and a category for offences related to indecency. Duplicate observations due to multiple imprisonment of recidivists are excluded from the analysis to avoid double counting. We categorise all prisoners according to the nature of their crime to be able to control for offence-specific effects in our analysis, reported below.¹⁷ Finally, the main analysis is restricted to a sub-group of individuals older than 17 years-of-age. Populations exposed to low biological standards of living usually reach their final adult height in their late-teen years – in extreme cases even later (see Eveleth and Tanner 1976). We add variables to control for any such catch-up growth in our samples.¹⁸

5. Empirical Strategy

The empirical strategy used to test for a “famine effect” relies on a test of differences in final average male height between the Famine-born cohort and the cohorts born in the decades prior to and immediately after the Famine. This strategy aims to capture the effect of malnutrition and worsening disease environment experienced by the Famine-born cohort, i.e., stunted growth. Our working hypothesis is this effect should result in a height drop for individuals in locations where famine conditions were most severe, in rural Ireland (see Figure 3). Our alternative hypothesis is we find no effect, which implies there was “extreme selection”.

We use a broad definition of the Famine. While the main famine years are 1845 to 1852, we include the birth years 1844 and 1853 to account for famine exposure during infancy (1844 birth cohort) and in utero (1853 birth cohort). The following formula outlines the testing framework: where H denotes height of individual i who was born in county c at time t. All results are conditional on a time trend which tests for differences in height between the cohorts born before, during and after the Famine, indicated by vector T. Other control variables, denoted by vector X, include literacy, religion, birth county-fixed effects, and the type of crime committed. X also contains controls for conviction periods when selection into the prison sample might have differed from normal times. We run our analysis separately for each prison.

We use the aforementioned framework to test for a famine effect in two different OLS settings. Our first strategy uses the rationale of a structural break to test for differences in height between a sequences of birth cohorts. This is also known as calculating the simple cohort difference. Our second strategy is based on the idea of a difference-in-difference approach, where the rate of excess mortality in individual i’s county during the Famine proxies famine exposure, i.e., the level of malnutrition and disease environment. In this setup, mortality rates help to distinguish high excess mortality regions of birth, such as rural Munster and Connacht, from low mortality regions, such as parts of Ulster and the Greater Dublin area. In this latter framework we keep controls for time (T) to capture general time-specific effects, but add an interaction term (M_c × F_it) combining famine-related mortality (M_c) and information about whether an individual was born during the Famine (F_it). This term proxies for famine severity and is equal to the excess mortality (in per cent) in an individual’s birth county if this individual was born during the Famine, and zero otherwise. To obtain a total effect in this setup, we combine all mortality-related effects obtained in the regression analyses with the general height trend estimated using birth cohort dummies. In models which solely use birth period effects, this combined metric is equal to the difference in height between pre-Famine and Famine-born cohorts. For models based on a combined strategy of using birth period dummies together with excess mortality, we compute the aggregate effect by entering the observed excess mortality into Formula 2.

OLS may be inefficient if the errors are non-normal. Implicitly, conventional estimators test for an average famine effect which applies to all points on the height distribution equally (see Figure 5). However, it is reasonable to assume the Famine did not affect tall and short individuals equally. Indeed, elsewhere scholars have found short individuals tend to come from lower socioeconomic backgrounds, while for tall individuals the opposite is true (Steckel 1995). We address the presumption selective mortality and migration (S1 and S2 above) disproportionately affected some parts of the height distribution in two ways: by using a Tobit specification, and by conducting a quantile regression analysis.

• Download figure
• Open in new tab

Figure 5:

Height histograms, male prisoners

Sources: Authors’ calculations, using Clonmel and Dublin prison registers.

Tobit regression techniques address potential selection bias in the left-hand-side of the height distribution, choosing a cut-off point which is below, but close, to the analysed samples’ median height. If the Tobit specification leads to substantially different results compared to the OLS results, we can conclude selective mortality and migration disproportionately affects the left hand-side of height distribution. We use Formula 4, assuming the specification of individual height outlined in Formula 3. where H_L is censored from below, and is a latent variable specified as: Quantile regression is more robust to non-normal errors and outliers. This method allows us to consider the impact of a set of covariates on the entire height distribution.¹⁹ We use a set of quantile regression models in a 10-year famine window setting to test for differences in the magnitude of the Famine within the height distributions. The conditional quantile function is defined as the τ-th quantile Q_Y|X τ|x of the conditional distribution function F_Y|X y|x for the dependent variable height y given the explanatory variable x, is a binary variable identifying a Famine-born individual. We report results for Dublin and County Tipperary using the following quantiles to assess differences in Famine exposure of different height percentiles: τ = 10; τ = 25 ;τ₃ = 50; τ₄ = 75; and τ₅ = 90.

6. Principal Results

Table 4 reports the origin and measurement of the individuals in each sample used, and separates city-born from rural-born individuals to enable a precise analysis of the effects of famine in this historical setting.²⁰ Regression coefficients indicate the magnitude of the famine effect according to Formulas 1 and 2, the number of observations, and the coefficient of determination.²¹ In addition, we report famine-related mortality in the birth counties of the individuals in each sample, and the total estimated change in height due to exposure to the Famine. These results are shown on the bottom line of the table.

For males born in Dublin County, our results suggest a modest height decline throughout the 30-year period under observation. Height declined by 0.3 cm during the Famine, and by an additional 0.7 cm after the Famine (Table 4, Model 1). We compare this Dublin-born, Dublin-imprisoned group with another sample which consists of within-Ireland migrants. By contrast, for Irishmen born outside Leinster we do not find such an obvious effect. In Models 2 to 4, the results suggest a total famine effect of 0.2 to 0.3 cm, but these coefficients are not statistically significant. For the period after the Famine, we find the height of those born in locations closer to Dublin tended to decrease (Model 1), while height in areas most heavily affected by the Famine stagnated, or even increased (Models 2 to 4).

We use another sample, inmates of Clonmel Gaol in County Tipperary, to gain insights into height developments in rural Ireland (Table 4, Models 5 to 8). We separately assess the heights of those hailing from Tipperary and those born in areas of Ireland outside the province of Munster. As for Tipperary’s urban-born individuals, we find an increase of 0.7 cm for the Famine-born cohort, and a further increase of 0.6 cm afterwards (Model 5). While mortality figures imply a significant famine effect, there was no corresponding drop in height; indeed, combining the mortality and height coefficients implies a stagnation or even an increase in height for the Famine-born generation. If the Famine left its mark on the Irish population, we expect to find signs of stunting in this rural sample of Munster-born prisoners. But here, results suggest heights were either unaffected (Models 6 and 9), or exhibited an increase (Model 8) in height during the Famine – although these coefficients are not statistically significant. Again, this finding is consistent with selection rather than scarring.

A set of quantile regressions, reported in Table 5, help us to further illustrate the relative roles of selection and scarring. They allow us to test whether this average effect can also be observed for each denoted percentile, or, indeed, whether the famine effect varies across the height distribution. Recall that for Dublin-born individuals, our OLS estimates suggest a minimal decrease in average height (as shown again in Table 5, Model 1). The Greater Dublin area was only modestly exposed to the Famine. Mortality was modest, but a general increase in food prices may have permanently affected the health of lower classes. For the 10th percentile, we observe an increase in height by approximately 1.0 cm, suggesting selection more than outweighed scarring. At the 25th, 50th and 75th percentiles we observe a conventional famine effect in that heights dropped by 0.5 cm, 1.1 cm and 0.4 cm. Interestingly, heights decreased more at the median than at the 25th or 75th percentile. We interpret this as evidence of the changing role of selection and scarring across the distribution. We do not observe a statistically significant change in height at the 90th percentile, suggesting the Famine had an effect which was minimal or absent among this class.

Recall that results for Clonmel suggest selection more than outweighs scarring, resulting in an increase in height of the famine cohort (as shown again in Table 5, Model 7). If the effect of selection (relative to scarring) is largest in locations which were most exposed to the Famine, then it is reasonable to expect this effect to be most pronounced among low percentiles. We find this positive effect was indeed strongest when using the 10th percentile as a benchmark, where we observe a height increase of 1.3 cm for the Famine-born cohort in County Tipperary (Table 5, Model 8). Indeed, quantile regression results point towards a famine effect similar to that observed in Dublin among the lowest classes. This effect is somewhat weaker for the 25th and 75th percentiles, ranging between approximately 0.8 and 1.0 cm. We find a modest positive effect for the median and 90th percentile, but these coefficients were not statistically significant.

7. Additional Results

We run a series of additional analyses in order to ascertain the robustness of our findings. These are reported in the Appendices. In the first exercise, reported in Appendix A, we adopt the same identification strategy as in previous tables, but: (1) we take a variety of different sub-samples from our dataset with different characteristics; and (2) we redefine our core famine period as running from 1845 to 1849 and adopt a shorter, 5-year, observation window for our difference-in-difference analysis. Our results are consistent with our main findings.

The second exercise, reported in Appendix B, uses an instrumental variable approach to address potential omitted variable bias resulting from unobserved individual characteristics. We use the number of so-called “fever hospitals” per 10,000 families in an individual’s county of origin in the era before the Famine to proxy a county’s capacity to address famine fever during the Famine. Fever hospitals were exogenous to famine exposure and correlate with our mortality variable in the first stage of a 2SLS setup. For Dublin, IV results generally confirm earlier findings. Individuals born during the Famine in Leinster were approximately 1.0 cm shorter; results for Irish immigrants born outside the region of Leinster suggest height stagnation. While we cannot confirm height increases for the Munster-born sub-sample, IV results are in line with Tobit regressions and indicate height stagnation. In general, however, results are consistent with our earlier findings for Clonmel Gaol: the absence of scarring, suggesting extreme selection.

The third exercise, reported in Appendix C, exploits the religious affiliation and literacy status of prison inmates to disentangle whether scarring and selection may be cancelling one another out. Contrary to economic intuition, we find Catholics and illiterates in the surviving population were less affected by the Famine, not more, despite their somewhat lower socioeconomic status.

In the fourth exercise, described in Appendix D, we re-weight our prison sample to better reflect the attributes of the general population, taken from census summaries compiled for the British Parliament. The new results are consistent with our baseline results. The evidence presented so far not only suggests the famine effect differs by location, but that this effect varies according to socioeconomic status within each location.

The fifth robustness exercise, in Appendix E, further redefines our birth cohort window length to track year-on-year height differences. However, to do so we must remove those individuals born in round years due to a phenomenon known as “age heaping”. Again, our baseline results are confirmed by this analysis in that we find no famine effect for rural Ireland, but rather a drop in height among the urban Famine-born cohorts.

A sixth exercise, Appendix F, attempts to account for possible famine stunting experienced during adolescence. We re-run our analysis with new samples of individuals who were 3-7, 8-12 and 13-17 at the end of the Famine. Much like for infants, we find a famine adolescence effect for Dublin-born individuals, but no such effect for those born elsewhere.

Finally, the seventh exercise, Appendix G, looks for evidence of scarring amongst Irish emigrants. We use a sample extracted from Fogel et al. (1990) containing the 2,977 Union Army soldiers above the age of 19 who were born in Ireland. We then adopt the same methodology as we describe in Section 6: we run an OLS regression of height (in cm) using five-year birth cohorts. We find the same height penaly pattern among Irish-born Americans fighting in the Union Army as Dublin prisoners.

8. Discussion and Conclusion

We adopted a research design which allows us to isolate a large sample of individuals born during the period of Ireland’s reduced food availability, and then use anthropometric indicators as a proxy for their biological standard of living. Our results are somewhat surprising. We find a positive correlation between famine severity and the health of the populations most exposed to famine-induced hunger and disease during infancy; low famine-related mortality is associated with the most severe drops in average adult height. How can an incidence of famine have no long-run health consequences for some survivors? Our research, in which we use unique and extensive hand-collected data on populations in different geographic locations across one famine episode, suggests the answer lies in the severity of the famine under study. We argue those individuals most affected by Ireland’s nineteenth-century famine underwent an “extreme selection”; those most exposed did not themselves survive in Ireland long enough to have their measurements taken; those who died, or migrated, do not, by definition, enter our Irish based samples.

The Great Irish Famine has left an indelible mark on the Irish psyche. It remains the key watershed in Irish economic and social history. It is still invoked in political discourse. And it has undoubtedly had a lasting effect on the island’s culture. But what apparently is far less obvious is its impact on the health and wellbeing of Ireland’s surviving population, those who did not themselves succumb to starvation and disease, and who stayed on the island rather than moved abroad. Yes, the Famine permanently changed the diet of the entire island. But we find only famine survivors hailing from those parts of the country which were least affected by the immediate effects of famine experienced any detectable permanent health consequences. From this we surmise that the scarring effect outweighs the selection effect only in Ireland’s urban centres; mortality did not select the population as much in these areas as most people survived into adulthood, including society’s least healthy.

We attribute the mechanism underpinning the selection and scarring outcomes to the demographic features of the Famine: the geographic regions and segments of society undergoing extreme selection had higher excess mortality, greater decreases in birth rates and higher cohort depletion rates. Using insights from our quantile regressions, the Famine likely disproportionately affected the poorest elements of society, those at the low end of the height distribution, while selection was minimal among Ireland’s better-off. Selective mortality (which we refer to as S1 in Section 3) was likely compounded by selective migration (S2); archaeological evidence of Famine-burials in Ireland were disproportionately young providing evidence for S1, while Famine-born Irish buried in London, for example, suggests Ireland exported its bad health (S2). Indeed, the view that Ireland exported its bad health is supported by our analysis of Union Army records from the US Civil War, where we find that soldiers born in Ireland during the Famine experienced a 2.34 cm height penalty compared to their older compatriots (see Appendix G).

The Famine resulted in, or at the very least accelerated the process of, economic change across Ireland. Post-Famine Ireland was in many respects a very different place than Ireland before the potato was blighted. The elimination through death or migration of vast swathes of society, and the collective realisation that the reliance on a single crop can have devastating consequences, permanently changed the structure of Irish agriculture, not to mention the nation’s eating habits. Land reforms precipitated by, and even instigated as a direct result of, the Famine accelerated the process of urbanisation and helped to provide industrialising cities with cheap labour inputs. But for those who grew up in the worst-affected areas and lived long enough to tell the tale of the Great Irish Famine to the next generation, its effect on their own personal physical health was apparently relatively benign.

We recommend famine severity and its associated selective mortality effect be included more explicitly in any future studies of catastrophic risks and health. Less severe famines, like other in utero and early childhood health shocks, can have long-lasting scarring effects on survivors. But more severe famines, like the Great Irish Famine experienced in rural areas, can act as Malthusian catastrophes in that they eliminate society’s most vulnerable and leave behind only its fittest. Without accounting for severity, anthropometric studies of the wellbeing of the survivors of natural or man-made disasters cannot draw meaningful inferences about their long-run impact on society.

Government Publications

British Parliamentary Papers [BPP] (1823). Report of inspectors general on the general state of the prisons of Ireland. (342).

____ (1826). Fourth report of the inspectors general on the general state of the prisons of Ireland. (173).

____ (1840). Eighteenth report of the inspectors general on the general state of the prisons of Ireland, 1839. [240].

____ (1843). Twenty-first report of the inspectors-general on the general state of the prisons of Ireland, 1842. [462].

____ (1847-48). Twenty-sixth report of the inspectors-general on the general state of the prisons of Ireland, 1847. [952].

____ (1849). Twenty-seventh report of the Inspectors-General on the general state of the prisons of Ireland, 1848. [1069].

____ (1851). Abstract of census of Ireland, 1841 and 1851. (673).

____ (1865). Judicial statistics, 1864, Ireland. [3563].

____ (1867). Forty-fifth report of the inspectors-general on the general state of the prisons of Ireland, 1866. [3915].

____ (1867-68). Judicial statistics, 1867, Ireland. [4071].

____ (1868-69). First annual report of the registrar-general of marriages, births, and deaths in Ireland. [4137]

____ (1871). Criminal and Judicial Statistics, 1871. [C.674].

____ (1873). Fifty-first report of the Inspectors-General on the general state of the prisons of Ireland, 1872 .[C. 837].

____ (1873). Criminal and judicial statistics, 1872, Ireland. [C.851].

____ (1874). Fifty-second report of the Inspectors-General on the general state of the prisons of Ireland, 1873. [C. 966].

____ (1874). Eighth annual report of the Registrar-General of the marriages, births, and deaths in Ireland. [C.968].

____ (1874). Criminal and judicial statistics, 1873, Ireland. [C.1034].

____ (1875). Fifty-third report of the inspectors-general on the general state of the prisons of Ireland, 1874. [C.1256].

____ (1877). Drunkenness (Ireland). (124).

____ (1878-79). First report of the General Prisons Board, Ireland, 1879; with appendices. [C.2447]

____ (1882). Criminal and judicial statistics, 1881, Ireland. [C.3355].

____ (1884). Summary proceedings for drunkenness in 1883 (Ireland). (1).

____ (1888). Copy of Rules and Regulations in Force in the Prisons in Ireland. (329).

____ (1889). Return of the Cost of Police. (350).

____ (1890). Twelfth report of the General Prisons Board, Ireland, 1889-90. [C.6182].

____ (1892). Criminal and judicial statistics, Ireland, 1891. [C.6782].

____ (1900). Twenty-second report of the General Prisons Board, Ireland, 1899-1900. [Cd.293].

____ (1902). Judicial statistics, Ireland, 1901. [Cd. 1208, 1187].

____ Department of Social Welfare [Republic of Ireland] (1955). Report of Commission on Emigration and Other Population Problems. pp. 1948–1954 R. 84 (Pr. 2541).