This method can be used to identify and deal with unmeasured confounding and other biases, e.g. selection bias
Kraemer et al ., 1997 ; Sedgwick, 2013 ; Arnold and Ercumen, 2016 .
Traditionally many of the exposures that we believe to be risk factors for psychiatric disorder have been implicated through observational designs. These infer causation from observations of association. However, association is not causation. Threats to causal inference include reverse causation, confounding, and selection bias (Rutter, 2007 ; Thapar and Rutter, 2015 ). For example, has the supposed risk factor of family discord arisen as the result of the individual's psychiatric disorder – reverse causation? Has the common factor of social disadvantage contributed to both the outcome of psychiatric disorder and family discord – confounding? Does cannabis have a causal risk effect on schizophrenia or is it that those with a higher propensity to develop schizophrenia are more likely to use cannabis – selection bias?
These are important threats because if they lead to misleading and inconsistent conclusions, this confuses clinicians, researchers, the general public, and patients. At worst it leads to wasted resources. The challenges in inferring causality are not just restricted to psychiatry. For example, observational studies suggested vitamin E had a protective effect on cardiovascular disease until randomized controlled trials (RCTs) suggested that this was not the case (Eidelman et al ., 2004 ). RCTs are often considered as the ‘gold standard’ for assessing causal effects. However, given that RCTs of many environmental exposures relevant to psychopathology are not going to be feasible or ethical, what is the alternative?
‘Natural experiments’ provide an alternative strategy. We refer to designs that take advantage of situations whereby links between the exposure and other variables are separated by naturally occurring events or situations. Unlike RCTs, the manipulation is not undertaken by the researcher. Some involve the design and others the statistical methods.
In this review, we will consider some types of natural experiments and describe how they have been applied in the field of psychiatry. The aim is not to provide an exhaustive account of different methods but rather to focus on the principles, design, and limitations. There are a number of other methods that in the interests of space will not be covered in this review but are discussed elsewhere (Rutter and Thapar, 2018 ). Although there are other reviews (e.g. Pingault et al ., 2018 ), we aim to describe a broad range of designs and will provide examples of findings that would be relevant to a clinician.
There is a growing trend toward viewing causal inference as a single approach based on considering what would have occurred if an individual had not been exposed to the risk? [see Krieger and Davey Smith ( 2016 ) for an excellent discussion]. However, we agree with Krieger and Davey Smith ( 2016 ) for taking a broader view; one that emphasizes convergence or ‘triangulation’ of findings across diverse types of designs that have different types of biases and assumptions. When the same finding is observed using different approaches, it provides greater confidence in inferring causality especially when such studies are conducted in different populations.
Many of the most important risk factors for psychopathology, such as life events and inter-personal discord, are person-dependent; they are not randomly allocated. Thus, it is unsurprising that decades of research have shown that many types of adversities run in families and are heritable (e.g. McGuffin et al ., 1988 ; Plomin, 2018 ). This raises the possibility that an association between exposure and psychiatric outcome could arise through familial or genetic confounding (Thapar and Rutter, 2009 , 2015 ).
It is for this reason that genetically informative designs such as twin studies have been invaluable for testing whether links between environmental exposures and psychopathology remain associated once genetic or familial confounds are taken into account.
Some designs, such as the discordant sib pair and in vitro fertilization (IVF) design (Thapar et al ., 2007 ), enable removal of genetic or familial confounds for prenatal exposures. For example, prenatal exposure to cigarette smoke has been linked with later risk for offspring attention-deficit/hyperactivity disorder (ADHD), conduct disorder, bipolar disorder, and schizophrenia. The effects could potentially be causal; for example, mediated by effects of nicotine on the developing brain. However, unmeasured confounds and selection biases are a concern, meaning that natural experiment designs have proved very useful here (Quinn et al ., 2017 ; Rice et al ., 2018 ).
Twin and adoption studies are not able to separate genetic confounds for prenatal exposures. That is because twins share their prenatal exposures and varying degrees of genetic liability and for adopted offspring, it is their biological mother who provides both the prenatal environmental and half of their genetic makeup. However, such designs are well-suited for assessing post-natal exposures. Some designs such as the children-of-twins design (D'Onofrio et al ., 2003 ) and adoption designs are especially well-suited for examining cross-generational environmental as well as genetic transmission (see Table 2 ).
Genetically informative designs and what they can be used to assess
Prenatal exposures | Postnatal exposures | Cross-generational transmission | |
---|---|---|---|
IVF design | + | + | + |
Maternal paternal exposure | + | ||
Discordant sib pair design | + | + | |
Twin design | + | ||
MZ twin discordance | + | ||
Children of twin design | + | + | + |
Adoption design | + | + |
One method that has been used to disaggregate intra-uterine and genetic or house-hold/familial-level influences involves testing associations between maternal v. paternal exposures during pregnancy and offspring outcomes ( Fig. 1 ). If the link is mediated by an intra-uterine effect, a stronger association would be expected for the maternal exposure. For example, in a UK population-birth cohort ALSPAC, strong associations were observed between maternal smoking in pregnancy and shorter birth length (Howe et al ., 2016 ) and lower birth weight in offspring (Langley et al ., 2012 ) that were not observed when exposure to paternal smoking was examined (see Table 1 ; this is an example of a negative control exposure). However, in this same cohort, associations between exposure to smoking in pregnancy and ADHD were as strong for maternal exposures as they were for paternal exposures even in the case of mothers who did not smoke. These results held when the contribution of additional passive smoking was considered. There are limitations to this design including the fact that parents will show similarities in exposures due to genetic (assortative mating) and social reasons and it is restricted to the sorts of exposures that both parents could feasibly experience in pregnancy.
Maternal v. paternal exposure.
Full biologic siblings share on average 50% of their genome. Thus differences between them can be used to assess family-level confounds that include genetic and shared environmental contributions.
(i) Prenatal exposures. As they share the same mother, they become of special interest when they have been differentially exposed to prenatal factors. For example, taking the example of maternal smoking in pregnancy and ADHD, eight studies of discordant sibling pairs have now found that the siblings who were unexposed to smoking in utero showed elevated levels of ADHD (Rice et al ., 2018 ). Similar findings were observed for conduct problems. Birth weight provided the negative control as the studies that examined this outcome found that the association with cigarette smoking remained strong. A recent, large discordant sibling study also failed to find support for a causal effect of exposure to prenatal smoking on severe mental illness (bipolar disorder and schizophrenia) suggesting the contribution of family-level confounders to previously observed associations (Quinn et al ., 2017 ). There are many limitations to this design that have been described elsewhere. These include the issue of selection as mothers are behaving differently in different pregnancies. For example, the sample consists of a group of mothers who are able to quit smoking in one pregnancy but not the other. Also, there is the problem that siblings will be born at different times and thus will be exposed to different family-level and population-level risks.
(ii) Assessing later adversities using a sibling pair design and its extension, the co-relative study. The discordant sibling pair design and its extension involving pairs of relatives from the same generation such as half-siblings and cousins have also been used to assess causal links between adolescent and adult exposures and psychiatric disorders. For example, the observed association between cannabis use and schizophrenia has been well-established. However, the causal relationship could be subject to question given that those who are at elevated familial or genetic liability or with prodromal symptoms could be more likely to use cannabis (confounding, selection bias, and reverse causation). In one large, Swedish study, the authors used an extended sibling pair design to investigate the causal relationship between cannabis and schizophrenia (Giordano et al ., 2015 ). The association was much attenuated once familial confounding was taken into account; the effect size also was diminished when potential prodromal effects were considered that was assessed by increasing the temporal delay between cannabis abuse and admission for schizophrenia (odds ratio 1.67). The findings suggested that there is a likely causal link between cannabis use and schizophrenia for some but that the effect size is not as strong as previously reported because of the contribution of familial confounding and reverse causation.
An alternative design that enables separation of prenatal exposures from genetic ones is based on individuals who have been conceived through assisted reproductive technologies. Some of these individuals are genetically related to the woman who undergoes the pregnancy and others are genetically unrelated (see Fig. 2 ). If a prenatal exposure has causal effects, then association with the offspring outcome should be observed regardless of whether mother–offspring dyads are genetically related or unrelated. That was the case for maternal smoking in pregnancy and lower birth weight (Thapar et al ., 2009 ) and also for associations between maternal reports of stress in pregnancy and lower birth weight and preterm birth (Rice et al ., 2010 ).
In-vitro fertilisation design.
However, for association between maternal smoking in pregnancy and a trait measure of ADHD in offspring (Thapar et al ., 2009 ) as well as conduct problems (Rice et al ., 2009 ), association was only observed in genetically related mother–offspring dyads not in the unrelated pairs, suggesting genetic confounding. The finding converges with those from the maternal v. paternal exposure and discordant sibling pair designs. Interestingly, the magnitude of association in the related pairs was similar to that observed in other observational studies and including measured confounders of the sort including in observational designs, such as parental psychopathology, social class did not remove the genetic confound.
That is, findings from this and other studies suggest that residual confounds remain a problem for observational studies and that including multiple confounders is not a substitute for an informative design.
The IVF design (Thapar et al ., 2007 ) has also been used to assess inter-generational transmission of psychopathology and to examine post-natal adversity. For example, using this approach, depression symptoms were found to be environmentally transmitted and environmental links were observed between hostile parenting and antisocial behavior in offspring (Harold et al ., 2011 ).
The IVF design does have a number of limitations however. These include the representativeness of the families who have undergone IVF treatment and the low prevalence of certain types of risk factors (e.g. maternal smoking in pregnancy).
Twin designs utilize the fact that monozygotic (MZ) twins share on average 100% of their genes (DNA sequence) and dizygotic (DZ) twins share on average 50% of their genome.
The twin design allows variation in any given measure to be partitioned into genetic and environmental variance. Where both exposure and psychiatric outcome are assessed, ideally longitudinally to avoid the problem of reverse causation, the association between exposure (e.g. life events) and outcome (e.g. depression) can be decomposed into genetic and environmental components. As the genetic covariance between exposure and outcome is explicitly modelled, essentially the genetic confound is removed. Here, the investigator is interested in whether there is an environmental link that remains between the exposure and outcome. This design has been invaluable in demonstrating a number of potentially causal environmental risk factors for psychopathology.
For example, family and twin studies of depression in childhood, adolescence, and adult life have observed a familial and genetic contribution to life events, mainly those that are person-dependent (e.g. losing a job) rather than ones that are independent (death of a relative), as well as to depression (McGuffin et al ., 1988 ; Plomin, 2018 ).
Twin studies that have investigated the link between life events and depression suggest that the association between independent life events and depression appears to be mainly or entirely environmental; that is consistent with a causal explanation (Kendler et al ., 1999 ). For dependent life events, there is a stronger genetic contribution to the link with depression. This seems to be partly explained by self-selection into risk exposure by those predisposed to depression (Kendler et al ., 1999 ) and becomes more prominent from adolescence onwards (Rice et al ., 2003 ).
Another example is the link between harsh parenting and antisocial behavior in children. One twin study found that the association with corporal punishment was primarily explained by genetic factors (Jaffee et al ., 2004 ). This could arise, for example, through parental response to the child's behavior which is genetically influenced. However, the findings for physical abuse were different. Here, the link with antisocial behavior was environmentally mediated, and consistent with a causal explanation.
Twin designs, their uses, strengths, and limitations have been described in detail elsewhere (State and Thapar, 2015 ). When genetic contributions are identified through bivariate twin analyses that we have described, it can index selection bias and potential threats to causal inference. However even with longitudinal twin designs, an environmentally mediated link does not prove a causal link between an exposure and outcome as there could be alternative pathways that explain the association including measurement artifacts.
This design utilizes the fact that MZ twins are considered to share 100% of their genes and means that differences in their phenotype are attributed to non-genetic contributions that include non-shared environment as well as measurement error and stochastic effects. The approach involves assessing whether MZ twins who are differentially exposed to a stressor or adversity (e.g. discordant for victimization) show differences in a given outcome (e.g. depression).
For example, in the UK E-risk twin study of 7–10 years old, 110 MZ twin pairs who were discordant for bullying victimization were assessed (Arseneault et al ., 2008 ). The co-twins who were bullied showed higher internalizing (anxiety/depression) symptom scores than those who were not exposed to bullying. A more recent US longitudinal twin study also investigated 145 MZ twins who were discordant for bullying victimization in childhood (Silberg et al ., 2016 ). Although being bullied showed a genetic link with social anxiety; there were also environmental links with social anxiety, separation anxiety, and young adult suicidal ideation. The findings from both of these studies are consistent with a causal effect of bullying victimization on emotional/anxiety symptoms and are important given the interest in reports from longitudinal observational designs.
In another longitudinal MZ discordant twin study (Caspi et al ., 2004 ), Caspi et al . assessed maternal hostility and warmth. This was achieved by conducting independent ratings from a recorded 5 min speech sample from the mother when talking about the child (expressed emotion EE). Maternal expressed emotion was found to be environmentally associated with later teacher-reported behavioral problems.
Although the MZ discordant pair design is useful because it controls for genetic confounding there are some drawbacks. For example, we now know that MZ twins are not per se 100% genetically identical, for example, through non-inherited genetic differences. Also discordant MZ twin pairs could be considered as atypical and rare especially for very highly heritable disorders such as autism or ADHD or schizophrenia. The exposure could be behaving as a proxy for some other risk factor that impacted on one twin and not the other.
The Children of Twins (CoT) design allow investigation of cross-generational links between parent and offspring psychopathology or parentally provided exposures and offspring outcomes. It takes advantage of the fact that the offspring of MZ and DZ twins are socially cousins (DZ twins are also genetically cousins) but the MZ twin offspring are genetically half siblings.
This type of design, for example, has been used to assess the cross-generational transmission of depression. In an Australian study of twins, their spouse, and offspring, environmental factors were found to explain the link between parents and offspring depression even when accounting for depression in spouses (Singh et al ., 2011 ). Similar findings had been found in an earlier US study (Silberg et al ., 2010 ). Another CoT study from Sweden found that depression symptoms in parents showed concurrent environmental but not genetic links with offspring internalizing symptoms (McAdams et al ., 2015 ). The findings accord with those from the IVF study (Harold et al ., 2011 ). A more recent Swedish CoT design observed only environmental transmission between parents and offspring for anxiety and neuroticism; again with no genetic contribution (Eley et al ., 2015 ).
These findings might appear puzzling in that while it is important to observe environmental transmission of depression and anxiety, there are no genetic contributions observed for either and this is inconsistent with twin studies (Sullivan et al ., 2000 ). Twin studies observe modest heritability for depression. One difficulty for cross-generational investigations is the assumption that the same genetic influences contribute across development when that is unlikely (e.g. Power et al ., 2017 ; Riglin et al ., 2017 ). Another issue is that twin study heritability estimates capture passive gene–environment correlation effects that would be reduced in CoT studies and eliminated in the IVF design.
The CoT design has also been used to assess postnatal adversities. One such study (Lynch et al ., 2006 ) found that harsh physical punishment remained associated with childhood behavioral problems even when genetic factors had been allowed for. These findings are in keeping with the twin study findings and taken together are consistent with harsh parenting having a causal effect on childhood antisocial behavior.
Adoption studies allow genetic and prenatal influences to be separated from post-adoption experiences. They provide a powerful method for assessing the contribution of rearing influences because these are known to be affected by with genetically influenced parental attributes. Ordinarily these biological parental characteristics would in turn be correlated with child characteristics including psychopathology thereby introducing a potential genetic confound. The advantage of adoption studies is that they remove this confound, the so-called passive gene–environment correlation because the genotypes of the parents who are rearing the children are independent of the child's genotypes.
There are several examples where adoption studies have been able to demonstrate the contribution of the rearing environment. For example, a study of adopted away children showed that negative parenting provided by the adoptive parent was associated with their adoptive child's antisocial behavior (Ge et al ., 1996 ). The adoptive parent's negative parenting was also associated with substance abuse/dependency or antisocial personality in the child's biological parents; that association appeared to be mediated via the child's behavior. Overall the findings suggested causal effects of negative parenting on children's antisocial behavior but also showed that the children's genetically influenced antisocial behavior in turn affected the parenting of the adoptive parents. The observation that negative parenting has a causal effect on offspring antisocial behavior converges with the findings from twin studies showing a convergence of findings from different designs.
A more recent example is provided by a Swedish large-scale adoption study cross-generational study (Kendler et al ., 2018 ). The authors were able to assess the contribution of genetic and rearing influences to parent–offspring resemblance for treated major depressive disorder. They found that both genetic and rearing influences contributed equally to parent–offspring resemblance in major depressive disorder. The adoptive families enabled the authors to further show that genetic and rearing influences acted additively rather than having an interactive effect. The authors highlighted that there had been four previous adoption studies of depression; although genetic contributions had previously been observed, only one had observed an environmental contribution to depression. However, now there have been two adoption studies that have showed an environmental contribution to inter-generational transmission of depression. Also the same findings have been observed in three children of twin designs and in the IVF design, although here some of these find environmental contributions only with no genetic transmission.
Overall the findings from different genetically informative studies of depression are converging on the suggestion that environmental/social factors contribute to the cross-generational transmission of depression. That of course has important clinical treatment and prevention implications.
Given a serious challenge to causal inference is selection or allocation bias, a number of studies have taken advantage of situations where risks have been introduced to or removed from an entire population.
Here, the best known studies are the Dutch Hunger Winter (Susser et al ., 1996 ) and Chinese famine studies (St Clair et al ., 2005 ) that examined the consequences of intra-uterine exposure to famine. These studies focused on populations that were exposed to universal time-limited famines that affected some individuals during the intra-uterine period. Exposed individuals in both studies showed around a twofold elevated risk of schizophrenia as well as congenital anomalies of the central nervous system. As there was no evidence for selection for exposure to either of these famines, the findings suggest that extreme nutritional deficiency in early pregnancy likely has a causal risk effect for schizophrenia. However, the conditions in both of these studies was extreme and atypical so whether the findings have relevance for the etiology of schizophrenia as a whole is unknown.
In this design, the strength is that it again removes selection or allocation bias whereby the person or some external agent influences the removal of risk. One good example is provided by the Great Smoky Mountains Study that is a longitudinal epidemiological study. During the course of this study of over 1000 children, a casino opened on a Native American reserve and provided a substantial increase to the family income for around a quarter of the original sample. The investigators were able to examine data before and after this happened. They showed that the relief of poverty led to decreased levels of oppositional defiant disorder and conduct disorder but not anxiety or depression (Costello et al ., 2003 ). The effects appeared to be mediated via altered parenting that included increased levels of supervision and parental time. Later follow-up showed that family income supplementation provided in childhood continued to be associated with lower rates of psychiatric problems including alcohol and cannabis abuse, lower rates of convictions for minor offenses, and higher levels of education. There were no links with later behavioral disorders or depression or other drug use (Costello et al ., 2010 ).
This design takes advantage of multiple waves of data that have been collected before and after the introduction or removal of the putative causal variable. This could be used to assess the impact of a policy or a naturally occurring event.
For example, after the introduction of UK legislation to reduce paracetamol package sizes, there was an observed drop in deaths from paracetamol overdoses (Hawton et al ., 2013 ).
Another example comes from a study of gang membership that is known to be associated with higher rates of delinquency (Thornberry et al ., 1993 ). However, it is not known whether that is due to selection effects with those having a propensity to be delinquent choosing to be in a gang or whether it is the causal social effects of being in a gang. Thornberry et al . ( 2002 ) found as might be expected important selection effects; boys who joined gangs were more delinquent than those who did not. However, they also showed that once boys left the gang, their rates of delinquency dropped off though not back to the level they were prior to joining the gang. This observation suggested that gang membership had additional social influences on delinquency. However, reverse causation and unmeasured confounders are possible contributors because we do not know what affected the boys’ decisions to leave the gang.
If these are applied to a whole nation and data are available before and after the introduction of the policy, then this can provide a useful natural experiment situation. One study in Sweden (Nilsson, 2008 ) focused on the effects of prenatal alcohol exposure in two regions that were subjected to an experimental policy change in alcohol sales. The intention was to shift the population away from drinking spirits to consuming drinks with a lower alcohol content. However, it inadvertently resulted in very marked increases in the consumption of strong beer especially amongst teenagers. The experimental policy started in 1967 but was terminated abruptly in mid-1968 once it was realized that alcohol consumption had increased. Using registry data, the researchers were able to assess a cohort of children who had been in utero during the exposed period. As the policy was time and geographically limited, the exposed cohort could be compared with unexposed cohorts in adjacent geographic regions and in adjacent time-unexposed cohorts. At around 30 years of age, the exposed group showed greatly reduced educational achievements, lower earnings, and greater welfare dependency than those born to the unexposed cohorts. The effects were strongest in males, those exposed for the longest in intrauterine life and those born in younger mothers. The results suggest that prenatal exposure to alcohol likely had intrauterine risk effects on offspring. However, the problem with this sort of policy study is that the results are obtained from analyses at a group rather than individual level.
One good example of a natural experiment was provided by the English and Romanian Adoptees Study that involved a very radical change in early environment. This is a longitudinal study of individuals who were exposed to institutional care and extreme privation from early infancy. The possibilities of selection bias and reverse causation were essentially removed because the children were admitted very early and virtually no children left care until the government regime fell in 1989. These children subsequently were exposed to a radical change in rearing environment after they were adopted into relatively advantaged homes in the UK. The findings from this study showed that although there was some recovery, early institutional care of the type experienced by these children for more than 6 months resulted in difficulties that persisted to adulthood including autistic-type symptoms, ADHD-like problems, disinhibited social engagement, and emotional symptoms but not cognitive impairment (Sonuga-Barke et al ., 2017 ).
As is the case for some of the other natural experiments, such as the famine studies, although selection bias is removed, the question is whether the findings apply to less severe and more common forms of deprivation.
An instrumental variable is a measured variable that is associated with the exposure of interest but that is not associated with the same selection effects and confounds. If the exposure has a genuinely causal risk effect on the outcome, then we would expect the instrumental variable also to be associated with the outcome. Early use and misuse of alcohol have been considered as potential causal risks or exposures for the later outcomes of alcohol dependence and misuse in adult life. Early puberty has been used as an instrumental variable for early use and misuse of alcohol because it is strongly associated with these exposures yet is not subject to the same selection biases or confounds.
Three studies have found that while early alcohol use and misuse in adolescence is associated with later alcohol problems, early puberty does not predict alcohol problems (Stattin and Magnusson, 1990 ; Caspi and Moffitt, 1991 ; Pulkkinen et al ., 2006 ).
These findings suggest that early alcohol use is likely an early manifestation of later alcohol problems rather than a cause of it.
Mendelian randomization (MR) utilizes the random assortment of parental genotypes to offspring during meiosis. Here a genetic variant that is robustly associated with the exposure is used as the instrumental variable and provided certain assumptions are met should provide a means of controlling for confounding and reverse causation (see Fig. 3 ). As more genetic variants are being identified through genome-wide association studies, there is a growing interest in using MR to test causal hypotheses and many methodological extensions of this approach (Davey Smith and Hemani, 2014 ). One approach called two-sample MR takes advantage of already published large genome-wide association studies. It uses genetic variants for exposures [e.g. C reactive protein (CRP)] as instrumental variables and another set of genetic variants from another independent GWAS for the outcomes variants (e.g. cardiovascular disease). MR has been used most successfully in relation to cardiovascular disease. For example, MR has been used to show that CRP does not have a causal risk effect on cardiovascular disease (C Reactive Protein Coronary Heart Disease Genetics Collaboration (CCGC) et al ., 2011 ). More recently, MR has started to be used in psychiatry; for example, a recent study observed body mass index effects on depression but not the reverse (Nagel et al ., 2018 ). MR is challenging because of its assumptions. For example, there is a need for genetic variants that have a strong and robust association with the exposure in question, although there are methods that allow for combining multiple genome-wide significant variants. Also if the genetic variant (instrument) has pleiotropic effects, and that is often the case, or influences a confounder or affects the outcome via another mechanism other than via the exposure, then that poses problems. There are methods for assessing pleiotropy and again, like all the methods we have discussed, MR findings on their own need to be interpreted with caution. However, when findings converge with other designs, they can be helpful in inferring causation. They are also a helpful alternative to RCTs.
Mendelian randomization. ( a ) The instrument is associated with the outcome only through the exposure. ( b ) Limitations – if the instrument is associated with a confounder or there is a horizontal pleiotropy.
It is crucial that genuinely causal influences on psychopathology are identified if interventions and policies are going to be effective. In recent years, findings relevant to psychiatry have emerged from different natural experiment designs and some are consistent across different designs; this strengthens causal inference. For example, hostile parenting affects antisocial behavior and RCTs uphold this causal inference. Genetically informative studies converge in favor of life events and victimization being environmentally linked with depression and environmental cross-generational transmission for depression. However, although smoking cessation programs for pregnant women are clearly a priority as cigarette smoke is detrimental to offspring physical health, the natural experiment designs suggest these will not be a useful means for preventing ADHD or antisocial behavior. So do natural experiments have an important future in the study of mental disorders? The answer is a firm yes.
AT receives grant funding from the Wellcome Trust and MRC.
3220 Accesses
14 Citations
34 Altmetric
Natural experiments or quasi-natural experiments in economics are serendipitous situations in which persons are assigned randomly to a treatment (or multiple treatments) and a control group, and outcomes are analysed for the purposes of putting a hypothesis to a severe test; they are also serendipitous situations where assignment to treatment ‘approximates’ randomized design or a well-controlled experiment.
This chapter was originally published in The New Palgrave Dictionary of Economics , 2nd edition, 2008. Edited by Steven N. Durlauf and Lawrence E. Blume
This is a preview of subscription content, log in via an institution to check access.
Institutional subscriptions
Ashenfelter, O., and A.B. Krueger. 1994. Estimates of the economic returns to schooling from a new sample of identical twins. American Economic Review 84: 1157–1173.
Google Scholar
Ashtekar, A., R.S. Cohen, D. Howard, J. Renn, S. Sarkear, and A. Shimony. 2003. Revisiting the foundations of relativistic physics: Festschrift in honor of john Stachel , Boston studies in the philosophy of science. Vol. 234. Dordrecht: Kluwer Academic.
Bastable, C.F. 1987. Experimental methods in economics (i). In The new palgrave: A dictionary of economics , ed. J. Eatwell, M. Milgate, and P. Newman, Vol. 2. London: Macmillan.
Bound, J., and G. Solon. 1999. Double trouble: On the value of twins-based estimation of the return to schooling. Economics of Education Review 18: 169–182.
Article Google Scholar
DiNardo, J. 2007. Interesting questions in freakonomics. Journal of Economic Literature .
DiNardo, J. and Lee, D.S.. 2002. The impact of unionization on establishment closure: A regression discontinuity analysis of representation elections. Working Paper No. 8993. Cambridge, MA: NBER.
DiNardo, J., and D.S. Lee. 2004. Economic impacts of new unionization on private sector employers: 1984–2001. Quarterly Journal of Economics 119: 1383–1441.
DiNardo, J., and T. Lemieux. 2001. Alcohol, marijuana, and American youth: The unintended consequences of government regulation. Journal of Health Economics 20: 991–1010.
Drake, S. 1981. Cause, experiment, and science: A galilean dialogue, incorporating a new english translation of Galileo’s bodies that Stay atop water, or move in it . Chicago: University of Chicago Press.
Fisher, R.A. 1935. Design of experiments . Edinburgh/London: Oliver & Boyd.
Hacking, I. 1983. Representing and intervening: Introductory topics in the philosophy of natural science . Cambridge: Cambridge University Press.
Book Google Scholar
Hacking, I. 1988. Telepathy: Origins of randomization in experimental design. Isis 79: 427–451.
Hacking, I. 2000. The social construction of what? Cambridge, MA: Harvard University Press.
Hearst, N., T.B. Newman, and S.B. Hulley. 1986. Delayed effects of the military draft on mortality: A randomized natural experiment. New England Journal of Medicine 314: 620–624.
Heckman, J.J. 2005. The scientific model of causality. Sociological Methodology 35: 1–97.
Heckman, J.J., and J.A. Smith. 1995. Assessing the case for social experiments. Journal of Economic Perspectives 9(2): 85–110.
Lee, D.S. 2008. Randomized experiments from non-random selection in U.S. house elections. Journal of Econometrics .
Magee, B. 2001. Talking philosophy: dialogues with fifteen leading philosphers . Oxford: Oxford University Press.
Mayo, D.G. 1996. Error and the growth of experimental knowledge science and its conceptual foundations . Chicago: University of Chicago Press.
Meyer, B. 1995. Natural and quasi-experiments in economics. Journal of Business and Economic Statistics 13: 151–161.
Morgan, M.S. 1987. Statistics without probability and Haavelmo’s revolution in econometrics. In The probabilistic revolution: Ideas in the sciences , ed. L. Krüger, G. Gigerenzer, and M.S. Morgan, Vol. 2. Cambridge, MA: MIT Press.
Nelson, A. 1990. Are economic kinds natural? In In Scientific Theories of Minnesota Studies in the Philosophy of Science , ed. C. Wade Savage, Vol. 14. Minneapolis: University of Minnesota Press.
Peirce, C.S.. 1958. In Collected Papers, vols. 7–8, ed. A. Burks. Cambridge, MA: Harvard University Press.
Rosenzweig, M.R., and K.I. Wolpin. 2000. Natural ‘natural experiments’ in economics. Journal of Economic Literature 38: 827–874.
Searle, J. 1995. The construction of social reality . New York: Free Press.
Shadish, W.R., T.D. Cook, and D.T. Campbell. 2002. Experimental and Quasi–Experimental designs for generalized causal inference . Boston: Houghton Mifflin.
Tribby, J. 1994. Club Medici: Natural experiment and the imagineering of ‘Tuscany’. Configurations 2: 215–235.
Voltaire. 1759. The history of candide; or all for the best , ed. C. Cooke. London, 1796.
Waller, R. 1684. Essayes of natural experiments made in the academie del cimento, under the protection of the most serene Prince Leopold of Tuscany . Facsimile edn, ed. R. Hall, trans. R. Waller. New York/London, 1964.
Wikipedia. 2006. Experiment. http://en.wikipedia.org . Accessed 28 Sept 2006.
Download references
Authors and affiliations.
http://link.springer.com/referencework/10.1057/978-1-349-95121-5
You can also search for this author in PubMed Google Scholar
Editors and affiliations, copyright information.
© 2008 The Author(s)
Cite this entry.
DiNardo, J. (2008). Natural Experiments and Quasi-Natural Experiments. In: The New Palgrave Dictionary of Economics. Palgrave Macmillan, London. https://doi.org/10.1057/978-1-349-95121-5_2006-1
DOI : https://doi.org/10.1057/978-1-349-95121-5_2006-1
Received : 12 September 2016
Accepted : 12 September 2016
Published : 15 December 2016
Publisher Name : Palgrave Macmillan, London
Online ISBN : 978-1-349-95121-5
eBook Packages : Springer Reference Economics and Finance Reference Module Humanities and Social Sciences Reference Module Business, Economics and Social Sciences
Policies and ethics
Natural Experiments in Management Research
The difference-in-difference (DID) design is probably the most frequently used design of natural experiments. The DID design aims to estimate an average treatment effect by comparing the pre-treatment to post-treatment changes in an outcome variable (e.g., job satisfaction, job performance) between a treatment group (i.e., individuals, groups or organizations that received a treatment) and a control group (i.e., individuals, groups or organizations that did not receive a treatment). The “simplest” form of the DID design is the two groups in two periods design ( Wing et al. 2018 ). For instance, company A may would like to know whether more delicious food in the company’s cafeteria is related to higher job satisfaction. To test the relationship, the company first collects data on employee job satisfaction at time t-0. Then, the company hires a renowed cook for its cafeteria in factory A (treatment group) but makes no changes in the cafeteria in factory B (control group). At time t-1, the company again collects data on employee job satisfaction. The results of the survey are as follows:
Factory A : Job satisfaction in t-0 = 4.2; job satisfaction in t-1 = 4.9 Factory B : Job satisfaction in t-0 = 4.5; job satisfaction in t-1 = 4.7
The figure below visualizes the changes in job satisfaction from t-0 to t-1 for both groups.
To calculate the average effect of the treatment (more delicious food) on the outcome (job satisfaction), we can use the rather simple formula:
Average treatment effect (ATE) = (yt-1,T – yt-0, T) – (yt-1, C – yt-0, C) where yt-1, T is the average job satisfaction of the treatment group in t-1 (4.9), yt-o, T is the average job satisfaction of the treatment group in t-0 (4.2), yt-1, C is the average job satisfaction of the control group in t-1 (4.7), and yt-0, C is the average job satisfaction of the control group in t-0 (4.5). Therefore, the ATE is (4.9-4.2) – (4.7-4.5) = 0.5, which indicates that more delicious food at the company’s cafeteria leads to an on average 0.5 points higher job satisfaction.
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Philip Ball was a physics editor of Nature during the Benveniste saga. His books include H 2 O: A Biography of Water and The Devil’s Doctor: Paracelsus and the World of Renaissance Magic and Science .
You can also search for this author in PubMed Google Scholar
The ‘natural experiments’ approach to economics that won three researchers the 2021 Sveriges Riksbank Prize in Economic Sciences has helped to make the field more robust, say economists.
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
24,99 € / 30 days
cancel any time
Subscribe to this journal
Receive 51 print issues and online access
185,98 € per year
only 3,65 € per issue
Rent or buy this article
Prices vary by article type
Prices may be subject to local taxes which are calculated during checkout
doi: https://doi.org/10.1038/d41586-021-02799-7
Angrist, J. D. & Krueger, A. B. Quart. J. Econ. 106 , 976–1014 (1991).
Article PubMed Google Scholar
Card, D. & Krueger, A. B. Am. Econ. Rev. 84 , 772–793 (1994).
Google Scholar
Imbens, G. W. & Angrist, J. D. Econometrica 62 , 467–475 (1994).
Download references
Reprints and permissions
Cincinnati Children’s Hospital seeks the next Director for the Division of Experimental Hematology and Cancer Biology.
Cincinnati, Ohio
Cincinnati Children's Hospital & Medical Center
Jointly sponsored by the Hangzhou Municipal People's Government and the University of Chinese Academy of Sciences.
Hangzhou, Zhejiang, China
Hangzhou Institute of Advanced Study, UCAS
Job Title: Associate or Senior Editor, Nature Energy Location: New York, Jersey City, Philadelphia or London — Hybrid Working Application Deadline:...
New York City, New York (US)
Springer Nature Ltd
IOP is the leading research institute in China in condensed matter physics and related fields. Through the steadfast efforts of generations of scie...
Beijing, China
Institute of Physics (IOP), Chinese Academy of Sciences (CAS)
FACULTY POSITION IN THE MOLECULAR BIOLOGY OF REPRODUCTION
Dallas, Texas (US)
The University of Texas Southwestern Medical Center (UT Southwestern Medical Center)
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
Quickonomics
### Definition of Natural Experiment
A natural experiment refers to an empirical or observational study in which the conditions for the research are determined by nature or by other factors outside the control of the investigators. Unlike controlled experiments where researchers manipulate variables to determine cause and effect, natural experiments take advantage of real-world situations where a natural division or variation occurs, allowing for the observation of its impact on specific outcomes. These types of experiments are particularly valuable in fields like economics, sociology, and epidemiology, where ethical or practical considerations make controlled experiments difficult to carry out.
### Example
Consider a situation where a certain city introduces a significant public transportation subsidy, effectively reducing the cost of using public transit, while a neighboring city does not. Over time, researchers might observe changes in traffic congestion, air quality, and public transit usage between the two cities. This scenario can be viewed as a natural experiment because the subsidy’s implementation creates a naturally occurring division, facilitating the study of its effects on urban transportation patterns without needing a controlled, experimental design.
### Why Natural Experiments Matter
Natural experiments are crucial for several reasons. They provide a unique opportunity to study the effects of variables that cannot be ethically or practically manipulated by researchers. By observing naturally occurring variations, scientists and policymakers can infer causal relationships in complex, real-world settings. This ability to glean insights from uncontrolled environments helps bridge the gap between theory and practice, offering evidence that can inform public policy, economic strategies, and social interventions. Furthermore, because they rely on real-world occurrences, the findings from natural experiments often have a high degree of external validity, meaning they can be generalized to broader contexts.
### Frequently Asked Questions (FAQ)
#### How do natural experiments differ from randomized controlled trials (RCTs)?
Natural experiments differ from RCTs in that the researcher has no control over the assignment of the treatment or intervention. In RCTs, subjects are randomly assigned to either the treatment or control group to ensure that any differences observed are due to the treatment itself and not to pre-existing conditions. In contrast, natural experiments rely on external circumstances to create an “experimental” and a “control” group, which may lead to challenges in isolating the treatment effect from other confounding factors.
#### What are the limitations of natural experiments?
One of the key limitations of natural experiments is the potential for confounding variables that researchers cannot control. These variables may influence the outcome of the experiment, making it difficult to establish causality with the same certainty as in controlled experiments. Additionally, since natural experiments are observational, they can be susceptible to selection bias, where the characteristics of the groups being compared differ in significant ways other than the treatment of interest.
#### Can natural experiments provide conclusive evidence on causality?
While natural experiments can provide valuable insights and suggest causal relationships, they may not offer conclusive evidence of causality. The lack of control over assignment and potential confounding factors means that results should be interpreted with caution. However, when designed and analyzed correctly, and ideally combined with other sources of evidence, natural experiments can contribute significantly to our understanding of causal mechanisms in real-world settings.
#### How are natural experiments identified and analyzed?
Identifying natural experiments often involves recognizing opportunities where natural or policy-related variations mimic the conditions of a controlled experiment. This can include changes in laws, technological advancements, or natural disasters. Analysis typically requires sophisticated statistical methods to account for potential confounding variables and to isolate the effect of the treatment or intervention. Techniques such as difference-in-differences, regression discontinuity designs, and instrumental variable analysis are commonly used to analyze data from natural experiments.
To provide the best experiences, we and our partners use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us and our partners to process personal data such as browsing behavior or unique IDs on this site and show (non-) personalized ads. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Click below to consent to the above or make granular choices. Your choices will be applied to this site only. You can change your settings at any time, including withdrawing your consent, by using the toggles on the Cookie Policy, or by clicking on the manage consent button at the bottom of the screen.
The independent and dependent variables are key to any scientific experiment, but how do you tell them apart? Here are the definitions of independent and dependent variables, examples of each type, and tips for telling them apart and graphing them.
The independent variable is the factor the researcher changes or controls in an experiment. It is called independent because it does not depend on any other variable. The independent variable may be called the “controlled variable” because it is the one that is changed or controlled. This is different from the “ control variable ,” which is variable that is held constant so it won’t influence the outcome of the experiment.
The dependent variable is the factor that changes in response to the independent variable. It is the variable that you measure in an experiment. The dependent variable may be called the “responding variable.”
Here are several examples of independent and dependent variables in experiments:
If you’re having trouble identifying the independent and dependent variable, here are a few ways to tell them apart. First, remember the dependent variable depends on the independent variable. It helps to write out the variables as an if-then or cause-and-effect sentence that shows the independent variable causes an effect on the dependent variable. If you mix up the variables, the sentence won’t make sense. Example : The amount of eat (independent variable) affects how much you weigh (dependent variable).
This makes sense, but if you write the sentence the other way, you can tell it’s incorrect: Example : How much you weigh affects how much you eat. (Well, it could make sense, but you can see it’s an entirely different experiment.) If-then statements also work: Example : If you change the color of light (independent variable), then it affects plant growth (dependent variable). Switching the variables makes no sense: Example : If plant growth rate changes, then it affects the color of light. Sometimes you don’t control either variable, like when you gather data to see if there is a relationship between two factors. This can make identifying the variables a bit trickier, but establishing a logical cause and effect relationship helps: Example : If you increase age (independent variable), then average salary increases (dependent variable). If you switch them, the statement doesn’t make sense: Example : If you increase salary, then age increases.
Plot or graph independent and dependent variables using the standard method. The independent variable is the x-axis, while the dependent variable is the y-axis. Remember the acronym DRY MIX to keep the variables straight: D = Dependent variable R = Responding variable/ Y = Graph on the y-axis or vertical axis M = Manipulated variable I = Independent variable X = Graph on the x-axis or horizontal axis
In “natural experiments,” the treatment or intervention is determined by variation not under the control of the researcher. These designs, used in economics and epidemiology to support inferences about causal relationships between interventions and outcomes, are useful tools to help improve the rigor of observational studies in health policy and medicine. Perhaps the first natural experiment in medicine was that of the English physician John Snow in the mid-nineteenth century. In 1854, a cholera outbreak struck Broad Street in London, killing hundreds. Studying case clusters, Snow discovered that neighborhoods supplied with water downstream of where sewage was discharged into the Thames River experienced high levels of disease, while neighborhoods receiving upstream water had low disease levels. 1 Snow described the populations as similar in age, occupation, income, and social rank, divided into groups without choice, illustrating an essential component of natural experiments: similar but distinct populations that are exposed to a condition outside the researchers’ control, allowing for reasonable conclusions about the potential causal link between exposure and outcome.
Randomized clinical trials (RCTs) have traditionally been viewed as the primary method for establishing causality in health care, but they have important limitations: they are expensive; it is not always possible to randomize patients; and their findings may not be generalizable to different patient populations or nonexperimental settings. When RCTs are not possible, medical and health policy researchers have turned to observational studies. In observational studies, however, individuals are not assigned to the intervention independently of potential confounding factors that could also influence outcomes, making it difficult to separate the treatment effect from other factors that may be associated with receiving the treatment.
By contrast, natural experiments rely on variation in treatment exposure that may be unrelated to other factors associated with the outcomes. Suppose researchers are interested in examining the likelihood of long-term use and adverse outcomes for patients after an initial opioid prescription. An observational analysis might be confounded if the factors that influence a clinician’s decision to prescribe opioids (eg, cancer-related pain) also affect long-term outcomes (eg, opioid dependence). An RCT might resolve this issue but would be ethically and practically challenging. Instead, researchers could examine how long-term opioid use varies among opioid-naive individuals who, by chance, are exposed to physicians with a high propensity vs low propensity to prescribe opioids (eg, when assigned to the next available physician in an emergency department). 2 In this scenario, the long-term outcomes following an initial opioid prescription could be identified by variation in the drug’s use associated with prescriber variation that is plausibly unrelated to variation in unobservable patient factors associated with both initial opioid use and long-term outcomes.
Natural experiments use quasi-randomization, a method of allocation to study groups that is not truly random and is not assigned by a researcher, such as a specific date, age, or event. These study designs have an important feature: the similarity of the groups can be measured. Treatment and control groups should be similar in sociodemographic characteristics, comorbidities, prior health care utilization, and any other factors that might be associated with outcomes, but often this is not the case and adjustments are needed based on these observed variables. Natural experiments attempt to control for unobserved variables. When well implemented, natural experiments may be more informative than traditional observational studies that do not control for unobservable confounders, but are less informative than RCTs in establishing true cause and effect. With natural experiments, the more closely the study design resembles an RCT, the more confidence we may have in the validity of the findings.
Five types of natural experiments are particularly relevant for observational studies in health policy and medicine: regression discontinuity designs (RDD), instrumental variable designs, difference-in-differences (DID) analyses, event-study analyses, and interrupted time-series 2 - 4 ( Table ). This is an overview of these types of studies with health policy examples and is not intended to provide a detailed assessment of these designs.
Regression discontinuity designs identify effect sizes associated with an intervention by studying individuals with treatment assignment that differs by position on either side of a specific, arbitrary cutoff (eg, a treatment threshold, policy implementation date, an age threshold, or a geographic discontinuity). 5 In this design, the probability of being exposed to the intervention changes discontinuously at this cutoff. Studies using RDD rely on the assumption that individuals on either side of the cutoff are similar, so their treatment assignment is nearly independent of their characteristics, both observed and unobserved. For example, a 2018 study evaluated the phased introduction of Medicare’s Value-Based Payment Modifier program. 3 Researchers used the program’s practice size thresholds (eg, 100 or more clinicians) to evaluate whether the program was associated with practice performance—with the assumption that practices just above and below the cutoff did not differ in important ways—and found that the program was not associated with improved practice performance and may have exacerbated health disparities.
Instrumental variable analyses using quasi-random variation in assignment to treatment or intervention have also been used to study clinical and health policy interventions. 6 For example, health policy researchers have been interested in whether higher spending hospitals achieve better outcomes, a relationship that is confounded by the fact that higher spending hospitals may treat patients that are disproportionately sicker, which could spuriously suggest that higher hospital spending leads to worse outcomes. To address this issue, a study examined the association between hospital spending and mortality by using quasi-random variation in ambulance dispatching patterns as an instrumental variable. 7 Ambulances may have preferences for which hospitals patients are taken to for reasons that are unrelated to patient clinical severity; this, in turn, may lead otherwise similar patients to be transported to (and treated at) higher vs lower spending hospitals.
Other natural experiments use different types of analyses to assess potential causal relationships. These include DID, event study, and interrupted time series analyses. In DID analysis, researchers compare outcomes in 2 groups that were similar before an intervention (natural or otherwise) that affected only 1 of the groups. 8 The DID analysis postulates that if the treatment had no effect, the differences between the groups would be unchanged after the treatment. One such study found lower long-term mortality rates after Hurricane Katrina among people who had been living in New Orleans compared with those who had been living in other similar cities, which they concluded represented the effect of migration because New Orleans residents migrated to areas with better socioeconomic conditions and lower baseline mortality after the hurricane. 9 A randomized experiment on the effects of resettling a population on that scale would have been infeasible.
In event-study analyses, researchers rely on exogenous and variable timing of interventions in exposed groups to study changes within groups over time (eg, estimating the effect size for the association between care continuity and outcomes by studying patients whose primary care physicians retired at different points in time). Although event-study analyses do not require control groups, control groups without any exposure are frequently incorporated into this approach. 4 Interrupted time series analyses are similar, but typically focus on changes in outcomes before and after a single event that affects a population of interest (eg, a citywide soda excise tax). 10
Each of these types of study designs and analyses have important limitations that should be considered, including not controlling for unobserved or unmeasured differences between the groups, risk of selection bias due to allocation that cannot be concealed from the researchers, non-parallel trends that could affect comparisons between the groups, and spillover influences from 1 group to the other. For example, in RDD studies, assumptions must be tested to ensure that observed variables are continuous at the point where the treatment and outcome discontinuities occur, such that there are no abrupt changes in the relationship between the observed variables and the treatment or outcome except at the discontinuity cutoff. Similarly, studies that use instrumental variable analysis must ensure that an appropriate instrumental variable is selected and should acknowledge the possible threats to validity from unmeasured confounding factors.
Natural experiments offer an important approach for examining potential causal links between interventions and outcomes. Studies that appropriately use these methods could help provide data to inform questions affecting the health of patients that otherwise may remain unanswered.
Published: June 11, 2021. doi:10.1001/jamahealthforum.2021.0290
Open Access: This is an open access article distributed under the terms of the CC-BY License . © 2021 Khullar D et al. JAMA Health Forum .
Corresponding Author: Anupam B. Jena, MD, PhD, Department of Health Care Policy, Harvard Medical School, 180 Longwood Ave, Boston, MA 02115 ( [email protected] ).
Conflict of Interest Disclosures : Dr Jena reports consulting fees from Pfizer, Bioverativ, Bristol Myers Squibb, Merck Sharp & Dohme, Janssen, Edwards Life Sciences, Novartis, Amgen, Eli Lilly, Vertex Pharmaceuticals, AstraZeneca, Celgene, Tesaro, Sanofi, Aventis, Precision Health Economics, and Analysis Group, all outside of the submitted work.
Khullar D , Jena AB. “Natural Experiments” in Health Care Research. JAMA Health Forum. 2021;2(6):e210290. doi:10.1001/jamahealthforum.2021.0290
© 2024
Artificial Intelligence Resource Center
Best of JAMA Network 2022
Browse and subscribe to JAMA Network podcasts!
COMMENTS
The authors note that an easy way to identify the independent or dependent variable in an experiment is: independent variables (IV) are what the researchers change or changes on its own, whereas ...
This table presents the results of the impact of agricultural tax reform on regional entrepreneurship. ENTRY is the natural logarithm of one plus a county level's total number of new registered enterprises during the year.SHOCK is the degree of fiscal stress change caused by the abolition of the agricultural taxes.POST is an indicator variable that equals to 1 for the year from 2005 to 2007 ...
To eliminate the interference of omitted variables on the research results, this article selects the following control variables 57,58: ... In a quasi-natural experiment, various factors may ...
Bell nonlocality is the resource that enables device-independent quantum information processing tasks. It is revealed through the violation of so-called Bell inequalities, indicating that the observed correlations cannot be reproduced by any local hidden-variable model. While well explored in few-body settings, the question of which Bell inequalities are best suited for a given task remains ...
Various metaheuristic algorithms inspired by nature have been designed to deal with a variety of practical optimization problems. As an excellent metaheuristic algorithm, the improved particle swarm optimization algorithm based on grouping (IPSO) has strong global search capabilities. However, it lacks a strong local search ability and the ability to solve constrained discrete optimization ...
This study examines data from 18 provinces in China over the period from 2012 to 2021, treating the introduction of prefabricated building policies as a quasi-natural experiment. Utilizing the difference-in-differences methodology, this research assesses the impact of these policies on the carbon emission performance of China's construction ...
Once you've signed up and been accepted to the EPP, you'll have access to a demo so you can experiment with this API. Join the early preview program. As of now, the Prompt API, summarization API, and the language detection API are available for prototyping.
A natural experiment is a study in which individuals ... First, both labor market outcomes and family size may be affected by unobserved "third" variables (e.g., personal preferences). Second, labor market outcomes themselves may affect family size (called "reverse causality"). For example, a woman may defer having a child if she gets a raise ...
Natural experiments (NEs) have a long history in public health research, stretching back to John Snow's classic study of London's cholera epidemics in the mid-nineteenth century. ... Directed acyclic graphs illustrating the assumptions of instrumental variable (IV) analysis. (a) The variable Z is associated with outcome Y only through its ...
Natural experiment, observational study in which an event or a situation that allows for the random or seemingly random assignment of study subjects to different groups is exploited to answer a particular question. Natural experiments are often used to study situations in which controlled ... In this way, at least one variable can be ...
1. What is a natural experiment? While the term natural experiment lacks an exclusively agreed upon definition, there is a general consensus that a natural experiment occurs when a particular intervention has been implemented but the circumstances surrounding the implementation are not under the control of researchers (Craig et al., Citation 2012, Citation 2011).
A natural experiment in psychology is a research method in which the experimenter observes the effects of a naturally occurring event or situation on the dependent variable without manipulating any variables. Natural experiments are conducted in the day (i.e., real life) environment of the participants, but here, the experimenter has no control ...
In practice, natural experiments are highly variable, and researchers face difficult choices about when to adopt a natural experimental approach and how best to exploit the opportunities that do occur. The value of a given natural experiment for research depends on a range of factors including the size of the population affected, the size and ...
A natural experiment is an empirical or observational study in which the control and experimental variables of interest are not artificially manipulated by researchers but instead are allowed to be influenced by nature or factors outside of the researchers' control. Unlike traditional randomized experiments, natural experiments are not ...
Natural Experiment. Experiments look for the effect that manipulated variables (independent variables, or IVs) have on measured variables (dependent variables, or DVs), i.e. causal effects. Natural experiments are studies where the experimenter cannot manipulate the IV, so the DV is simply measured and judged as the effect of an IV.
An IV approach to years of schooling. suspected to be endogenous The endogenous explanatory variable is. First stage (= 1 + reduced form regression): kk −1 + αα zz predicted using only exogenous information. ᬅ쟃 Exogenous variable = quarter-of-birth using age-based school enrollment laws.
Natural experiments help answer important questions This year's Laureates - David Card, Joshua Angrist and Guido Imbens - have shown that natural experiments can be used to answer central questions for society, such as how minimum wages and immigration affect the labour market. They have also clarified exactly which conclusions about cause
Population health interventions are essential to reduce health inequalities and tackle other public health priorities, but they are not always amenable to experimental manipulation. Natural experiment (NE) approaches are attracting growing interest as a way of providing evidence in such circumstances. One key challenge in evaluating NEs is selective exposure to the intervention. Studies should ...
Natural experiments provide an alternative strategy to randomized controlled trials as they take advantage of situations whereby links between exposure and other variables are separated by naturally occurring events or situations. In this review, we describe a growing range of different types of natural experiment and highlight that there is a ...
A natural experiment is a real world situation that resembles an experiment without any intervention or control by experimenters. The term is used for empirical studies based on variables and control groups that occur spontaneously. Natural experiments are rarely as well controlled as an experiment in a lab but allow for unique scale and scope of research.
Thus, when using natural experiments with as-if randomization, it is particularly important to consider potential confounding factors because the absence of random assignment can lead to confounding bias in standard natural experiments and violations of assumptions in instrumental-variable designs using a natural experiment.
Natural experiments or quasi-natural experiments in economics are serendipitous situations in which persons are assigned randomly to a treatment (or multiple treatments) and a control group, and outcomes are analysed for the purposes of putting a hypothesis to a severe test; they are also serendipitous situations where assignment to treatment 'approximates' randomized design or a well ...
forward, and that the strength of evidence derived from evaluating natural experiments is highly variable depending on the type of research design used in the evaluation. 1.2. Choosing when to use a natural experimental approach Considering that natural experiments occur in an ongoing basis creates a challenge for determining
The difference-in-difference (DID) design is probably the most frequently used design of natural experiments. The DID design aims to estimate an average treatment effect by comparing the pre-treatment to post-treatment changes in an outcome variable (e.g., job satisfaction, job performance) between a treatment group (i.e., individuals, groups or organizations that received a treatment) and a ...
In the natural sciences and medicine, researchers might seek to establish causal relationships by conducting experiments, such as RCTs, in which one variable is altered while others are kept the same.
Published Apr 29, 2024. ### Definition of Natural Experiment. A natural experiment refers to an empirical or observational study in which the conditions for the research are determined by nature or by other factors outside the control of the investigators. Unlike controlled experiments where researchers manipulate variables to determine cause ...
Here are several examples of independent and dependent variables in experiments: In a study to determine whether how long a student sleeps affects test scores, the independent variable is the length of time spent sleeping while the dependent variable is the test score. You want to know which brand of fertilizer is best for your plants.
Natural Experiments in the Social Sciences This unique book is the first comprehensive guide to the discovery, analysis, and ... 4.2 Instrumental-variables designs: natural experiments 92 4.2.1 Lotteries 94 4.2.2 Weather shocks 95 4.2.3 Historical or institutional variation induced by deaths 97 4.3 Conclusion 101
Five types of natural experiments are particularly relevant for observational studies in health policy and medicine: regression discontinuity designs (RDD), instrumental variable designs, difference-in-differences (DID) analyses, event-study analyses, and interrupted time-series 2 - 4 ( Table ). This is an overview of these types of studies ...