Correlation Does Not Imply Causation: Essays in Causal Inference

Nadja van 't Hoff

doi:10.21996/rrrr-pv69

Correlation Does Not Imply Causation: Essays in Causal Inference

Nadja van 't Hoff

Publikation: Afhandling › Ph.d.-afhandling

142 Downloads (Pure)

Abstract

Denne afhandling fokuserer på at forbedre politiske anbefalinger ved at levere både nye teoretiske og empiriske indsigter inden for feltet kausal inferens. De første to kapitler adresserer huller i metoden for instrumentvariabler, mens det tredje kapitel giver en interessant anvendelse af difference-in-difference (DID) metoden for at evaluere effekten af den danske fedtafgift. Kapitel 1 foreslår antagelsen om limited monotonicity (LiM), som generelt er løsere end andre monotonicitets-antagelser. Denne antagelser betyder, at der kan udledes en identification af et parameter vi kalder combined complier local average treatment effect (CC-LATE). Dette nye parameter, der dækker en stor complier-population, har intuitive vægte. Kapitel 2 omhandler identifikation af kausale effekter af ikke-binære, diskrete interventioner ved brug af flere instrumenter og inklusion af kovariater. Kapitlet præsenterer nye indsigter i two-stage least squares metoden og foreslår combined complier average causal response (CC-ACR), et parameter med en intuitiv fortolkning, under LiM-antagelsen. Kapitlet udvikler også en ny test til påvisning af lokale brud på LiM-antagelsen. Kapitel 3 anvender en dobbeltrobust DID-metode til at undersøge de kausale effekter af den danske fedtafgift på forbrugeradfærd. Kapitlet finder betydelig effekt-heterogenitet på tværs af produkter, hvor forbruget af nogle produkter falder i den beskattede periode, og udgifterne til andre stiger. Den korte implementering af afgiften førte til en overordnet stigning i forbruget af visse produkter efter afgiftsperioden og rejser bekymringer om utilsigtede konsekvenser af afgiften. Samlet set bidrager afhandlingen betydeligt til områderne kausal inferens og folkesundhed. CC-LATE- og CC-ACR-parametrene er værdifulde værktøjer både for forskere og beslutningstagere, og resultaterne vedrørende den danske fedtafgift giver vigtige indsigter for udformningen af effektive politikker til fremme af en sund ernæring.

Originalsprog	Engelsk
Bevilgende institution	Syddansk Universitet
Vejledere/rådgivere	Mellace, Giovanni, Hovedvejleder Dahl, Christian Møller, Bivejleder
Udgiver	Syddansk Universitet. Det Samfundsvidenskabelige Fakultet
DOI	https://doi.org/10.21996/rrrr-pv69
Status	Udgivet - 23. jan. 2024

Dokumenter og links

10.21996/rrrr-pv69

Open Access VersionForlagets udgivne version, 15,4 MB

SDU link

Tjek adgangen til materialet i Syddansk Universitetsbiblioteks søgemaskine

Citationsformater

@misc{d59c4d273b344b6995560b7f74256f30,

title = "Correlation Does Not Imply Causation: Essays in Causal Inference",

abstract = "The present thesis focuses on improving policy recommendations by providing both newfound theoretical and empirical insights within the field of causal inference. The first two chapters address gaps in the instrumental variable methodology, while the third chapter provides an interesting application of the difference-in-difference (DID) approach to evaluate the effectiveness of the Danish fat tax. Chapter 1 proposes the limited monotonicity (LiM) assumption, which is generally weaker than other forms of monotonicity. Under this assumption, it provides an identification result for the combined complier local average treatment effect (CC-LATE), a novel parameter for a large complier population that has intuitive weights. Chapter 2 considers identification of causal effects of nonbinary, discrete treatments using multiple instruments and including covariates. It presents novel insights into the two-stage least squares methodology and proposes the combined complier average causal response (CC-ACR), a parameter with an intuitive interpretation, under the LiM assumption. It also develops a novel test for detecting local violations of LiM. Chapter 3 employs a doubly-robust DID approach to examine the causal effects of the Danish fat tax on consumer behavior. It finds significant effect heterogeneity across products, with consumption of some products declining during the taxed period and expenditure on others increasing. The brief implementation of the tax led to an overall increase in consumption for certain products in the post-tax period, raising concerns about unintended consequences of the tax. Overall, the thesis makes significant contributions to the fields of causal inference and public health policy. The CC-LATE and CC-ACR parameters are valuable tools for researchers and policymakers alike, and the findings on the Danish fat tax provide important insights for designing effective nutrition policies.",

author = "{van 't Hoff}, Nadja",

year = "2024",

month = jan,

day = "23",

doi = "10.21996/rrrr-pv69",

language = "English",

publisher = "Syddansk Universitet. Det Samfundsvidenskabelige Fakultet",

school = "SDU",

}

TY - GEN

T1 - Correlation Does Not Imply Causation: Essays in Causal Inference

AU - van 't Hoff, Nadja

PY - 2024/1/23

Y1 - 2024/1/23

N2 - The present thesis focuses on improving policy recommendations by providing both newfound theoretical and empirical insights within the field of causal inference. The first two chapters address gaps in the instrumental variable methodology, while the third chapter provides an interesting application of the difference-in-difference (DID) approach to evaluate the effectiveness of the Danish fat tax. Chapter 1 proposes the limited monotonicity (LiM) assumption, which is generally weaker than other forms of monotonicity. Under this assumption, it provides an identification result for the combined complier local average treatment effect (CC-LATE), a novel parameter for a large complier population that has intuitive weights. Chapter 2 considers identification of causal effects of nonbinary, discrete treatments using multiple instruments and including covariates. It presents novel insights into the two-stage least squares methodology and proposes the combined complier average causal response (CC-ACR), a parameter with an intuitive interpretation, under the LiM assumption. It also develops a novel test for detecting local violations of LiM. Chapter 3 employs a doubly-robust DID approach to examine the causal effects of the Danish fat tax on consumer behavior. It finds significant effect heterogeneity across products, with consumption of some products declining during the taxed period and expenditure on others increasing. The brief implementation of the tax led to an overall increase in consumption for certain products in the post-tax period, raising concerns about unintended consequences of the tax. Overall, the thesis makes significant contributions to the fields of causal inference and public health policy. The CC-LATE and CC-ACR parameters are valuable tools for researchers and policymakers alike, and the findings on the Danish fat tax provide important insights for designing effective nutrition policies.

AB - The present thesis focuses on improving policy recommendations by providing both newfound theoretical and empirical insights within the field of causal inference. The first two chapters address gaps in the instrumental variable methodology, while the third chapter provides an interesting application of the difference-in-difference (DID) approach to evaluate the effectiveness of the Danish fat tax. Chapter 1 proposes the limited monotonicity (LiM) assumption, which is generally weaker than other forms of monotonicity. Under this assumption, it provides an identification result for the combined complier local average treatment effect (CC-LATE), a novel parameter for a large complier population that has intuitive weights. Chapter 2 considers identification of causal effects of nonbinary, discrete treatments using multiple instruments and including covariates. It presents novel insights into the two-stage least squares methodology and proposes the combined complier average causal response (CC-ACR), a parameter with an intuitive interpretation, under the LiM assumption. It also develops a novel test for detecting local violations of LiM. Chapter 3 employs a doubly-robust DID approach to examine the causal effects of the Danish fat tax on consumer behavior. It finds significant effect heterogeneity across products, with consumption of some products declining during the taxed period and expenditure on others increasing. The brief implementation of the tax led to an overall increase in consumption for certain products in the post-tax period, raising concerns about unintended consequences of the tax. Overall, the thesis makes significant contributions to the fields of causal inference and public health policy. The CC-LATE and CC-ACR parameters are valuable tools for researchers and policymakers alike, and the findings on the Danish fat tax provide important insights for designing effective nutrition policies.

U2 - 10.21996/rrrr-pv69

DO - 10.21996/rrrr-pv69

M3 - Ph.D. thesis

PB - Syddansk Universitet. Det Samfundsvidenskabelige Fakultet

ER -

Correlation Does Not Imply Causation: Essays in Causal Inference

Abstract

Dokumenter og links

SDU link

Fingeraftryk

Citationsformater