Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Causal Inference in Observational Studies and Experiments: Theory and Applications
Uppsala University, Disciplinary Domain of Humanities and Social Sciences, Faculty of Social Sciences, Department of Statistics.ORCID iD: 0000-0002-1260-7737
2019 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

This thesis consists of six papers that study the design of observational studies and experiments.

Paper I proposes strategies to consistently estimate the average treatment effect of the treated using information derived from a large number of pre-treatment measurements of the outcome. The key to this strategy is to use two-level time-series model estimates to summarize the inter-unit heterogeneity in the sample. It is illustrated how this approach is in line with the conventional identifying assumptions, and how sensitivity analyses of several key assumptions can be performed.

Paper II contains an empirical application of the identification strategy proposed in Paper I. This study provides the first causal analysis of the demand response effects of a billing demand charge involuntarily introduced to small and medium sized electricity users.

Paper III proposes strategies for rerandomization. First, we propose a two-stage allocation sample scheme for randomization inference to the units in balanced experiments that guarantees that the difference-in-means estimator is an unbiased estimator of the sample average treatment effect for any experiment, conserves the exactness of randomization inference, and halves the time consumption of the rerandomization design. Second, we propose a rank-based covariate-balance measure which can take into account the estimated relative weight of each covariate.

Paper IV discusses the concept of optimal rerandomization. It is shown that depending on whether inference is to be drawn to the units of the sample or the population, the notion of optimal differs. We show that it is often advisable to aim for a design that is optimal for inference to the units of the sample, as such a design is often near-optimal also for inference to the units of the population.

Paper V summarizes the current knowledge on asymptotic inference for rerandomization designs and proposes some simplifications for practical applications. Drawing on previous work, we show that the non-normal sampling distribution of the difference-in-means test statistic approaches normal as the rerandomization criterion approaches zero. Furthermore, the difference between the correct non-normal distribution and the proposed approximation based on a normal distribution is in many situations negligible even for near optimal rerandomization criteria.

Paper VI investigates and clarifies the relation between the traditional blocked designs and rerandomization. We show that blocking and rerandomization is very similar, and in some special cases identical. Moreover, it is shown that combining blocking and rerandomization is always at least as efficient as using only rerandomization, but the difference is in many cases small.

Place, publisher, year, edition, pages
Uppsala: Acta Universitatis Upsaliensis, 2019. , p. 36
Series
Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Social Sciences, ISSN 1652-9030 ; 172
Keywords [en]
experimental design, identification, observational studies, rerandomization
National Category
Probability Theory and Statistics
Research subject
Statistics
Identifiers
URN: urn:nbn:se:uu:diva-393810ISBN: 978-91-513-0762-6 (print)OAI: oai:DiVA.org:uu-393810DiVA, id: diva2:1355365
Public defence
2019-12-06, Hörsal 2, Ekonomikum, Kyrkogårdsgatan 10, Uppsala, 10:15 (English)
Opponent
Supervisors
Available from: 2019-11-05 Created: 2019-09-27 Last updated: 2019-11-27
List of papers
1. Using high frequency pre-treatment outcomes to identify causal effects in non-experimental data: Causal effects in non-experimental data
Open this publication in new window or tab >>Using high frequency pre-treatment outcomes to identify causal effects in non-experimental data: Causal effects in non-experimental data
2018 (English)Report (Other academic)
Abstract [en]

In observational studies it is common to use matching strategies to consistently estimate the average treatment effect of the treated (ATET) under the unconfoundedness assumption of the outcome and the treatment assignment mechanism. Matching is often based on a set of time invariant covariates together with one or a few pre-treatment measurements of the outcome. This paper proposes estimation strategies using a large number of pre-treatment measurements of the outcome to consistently estimate the average treatment effect of the treated (ATET). The assumptions under which these approaches are valid are given. It is shown when and how the strategies can be used to replace, or add to, time-invariant covariates to identify and consistently estimate the ATET. The theoretical results and estimation strategies are illustrated by a study of electricity consumption.

Publisher
p. 34
Series
Working paper / Department of Statistics, Uppsala University ; 2018:1
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:uu:diva-355260 (URN)
Note

A revised version with a new title was published 2019-01-25.

A second revised version was published 2019-09-30.

Available from: 2018-06-27 Created: 2018-06-27 Last updated: 2019-09-30Bibliographically approved
2. Identifying and estimating the effects of a mandatory billing demand charge
Open this publication in new window or tab >>Identifying and estimating the effects of a mandatory billing demand charge
2019 (English)In: Applied Energy, ISSN 0306-2619, E-ISSN 1872-9118, Vol. 237, p. 885-895Article in journal (Refereed) Published
Abstract [en]

As peak demand for electricity continues to rise, distributors have begun charging small and medium-sized users for their short term demand rather than just their energy use. This is not only to meet the political aspirations for increased demand-side flexibility that now exist in many corners of the world, but to make sure that users are charged for the costs they incur. As it is only until recently that this type of users have come to face demand charges, there are however very few studies on what the actual effects of such pricing policies are, and those studies that do exist suffer from different methodological shortcomings that reduce their validity as a basis for real-world policy evaluations. This study provides the first state-of-the-art causal analysis of the demand response effects of a billing demand charge involuntarily introduced to small and medium sized users (35–63 A), using novel two-level time series models on retrospective observational consumption and survey data. Our analyses suggest that the tariff has induced an average response of −0.32 kWh/day per user over a two year long posttreatment period in comparison to a matched control group, equal to 7.4% of their daily average use during the pretreatment period. The response seems to have increased over time and to be greater during wintertime: around −0.70 kWh/day or 16.2% of the treated users’ average daily use during the pretreatment period. Comparing the individual users’ response to the size of their financial incentive to respond given the new tariff as well as their self-reported perception of the relative importance of electricity expenditures, we did not find any support for the common assumption that users with a higher financial incentive to respond do so to a greater extent. This might suggest that small and medium-sized commercial users, just as residential users, may exhibit non-financial drivers and barriers for engaging in demand response that may be vital to understand as policy makers and industry continue to seek increased demand-side flexibility.

Keywords
Demand response; Demand-based; Capacity charge; Capacity-based; Causal inference
National Category
Other Social Sciences Other Engineering and Technologies
Identifiers
urn:nbn:se:uu:diva-374770 (URN)10.1016/j.apenergy.2019.01.028 (DOI)000459845100066 ()
Projects
Marknadsstyrd effekttariff inom eldistributionen
Funder
StandUp
Note

Projektet finansierades även av Energiforsk (EV32254).

Available from: 2019-01-24 Created: 2019-01-24 Last updated: 2019-09-27Bibliographically approved
3. Re-randomization strategies for balancing covariates using pre-experimental longitudinal data
Open this publication in new window or tab >>Re-randomization strategies for balancing covariates using pre-experimental longitudinal data
2019 (English)Report (Other academic)
Abstract [en]

This paper considers experimental design based on the strategy of rerandomization to increase the effciency in experiments. Two aspects of rerandomization are addressed. First, we propose a two-stage allocation sample scheme for randomization inference to the units of the experiments in balanced experiments that guarantees that the difference-in-mean estimator is an unbiased estimator of SATE for any experiment, conserves the exactness of randomization inference, and halves the time consumption of the rerandomization design. Second, we propose a rank-based covariate balance measure which can take into account the estimated relative weight of each covariate. Several strategies for estimating these weights using pre-experimental data are proposed. Using Monte Carlo simulations, the proposed strategies are compared to complete randomization and Mahalanobis-based rerandomization. An empirical example is given where the power of a mean difference test of the electricity consumption of 54 households is increased by 99%, in comparison to complete randomization, using one of the proposed designs based on high frequency longitudinal electricity consumption data.

Place, publisher, year, edition, pages
Department of Statistics, Uppsala University, 2019. p. 38
Series
Working paper / Department of Statistics, Uppsala University ; 2018:4
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:uu:diva-364864 (URN)
Note

Title of the first published version of the report was: Experimental design using longitudinal data

A revised version with a new title was published 2019-01-25.

A second revised version was published 2019-09-30.

Available from: 2018-11-05 Created: 2018-11-05 Last updated: 2019-09-30Bibliographically approved
4. On optimal re-randomization designs
Open this publication in new window or tab >>On optimal re-randomization designs
2019 (English)Report (Other academic)
Abstract [en]

Blocking is commonly used in randomized experiments to increase efficiency of estimation. A generalization of blocking is to remove allocations with imbalance in covariates between treated and control units, and thenrandomize within the set of allocations with balance in these covariates. This idea of rerandomization was formalized by [5], who suggested using the affinely invariant Mahalanobis distance between treated and control covariate means as the criterion for removing unbalanced allocations. [3] proposed reducing the set of balanced allocations to the minimum. Here we discuss the implication of such an ‘optimal’ rerandomization design for inferences to the units inthe sample and to the population from which the units in the sample were randomly drawn. We argue that, in general, it is a bad idea to seak the optimal design for an inference to the population because that inference typically only reflects uncertainty from the usually hypothetical random sampling, and not the randomization of treatment versus control.

Publisher
p. 20
Series
Working paper / Department of Statistics, Uppsala University ; 2019:3
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:uu:diva-381381 (URN)
Available from: 2019-04-09 Created: 2019-04-09 Last updated: 2019-09-27Bibliographically approved
5. Asymptotic Inference for Optimal Re-Randomization Designs
Open this publication in new window or tab >>Asymptotic Inference for Optimal Re-Randomization Designs
2019 (English)Report (Other academic)
Abstract [en]

Recently an experimental design strategy, called rerandomization, has been proposed as acomplement to traditional blocked designs. The idea of rerandomization is to remove, from consideration, those allocations with large imbalances in observed covariates according to abalance criterion, and then randomize within the set of acceptable allocations. This paper clarifies the concept of an ‘optimal’ rerandomization design for inferences using the mean difference sample estimator, SATE. Based on the Mahalanobis distance criterion for balancing the covariates, we show that standard asymptotic inference to the population, from which the units in the sample are randomly drawn, is possible using only the set of best, or ‘optimal’, allocations.

Place, publisher, year, edition, pages
Uppsala universitet, 2019. p. 17
Series
Working paper / Department of Statistics, Uppsala University ; 2019:2
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:uu:diva-381384 (URN)
Note

Title of the first published version of the report was: Optimal designs and asymptotic inference. A revised version with a new title was published 2019-09-30.

Available from: 2019-04-09 Created: 2019-04-09 Last updated: 2019-09-30Bibliographically approved
6. Re-randomization: A complement or substitute for stratification in randomized experiments?
Open this publication in new window or tab >>Re-randomization: A complement or substitute for stratification in randomized experiments?
2019 (English)Report (Other academic)
Abstract [en]

Rerandomization is a strategy for improving balance on observed covariates in randomized control trails. It has been proposed as a complement to traditional stratied (blocked) designs. However, the relationship and differences between stratification, rerandomization, and the combination of the two have not been previouslyinvestigated. In this paper, we show that stratified designs can be recreated by rerandomization and explain why, in most cases, stratification on binary covariates followed by rerandomization on continuous covariates is more efficient than rerandomization on all covariates at the same time.

Place, publisher, year, edition, pages
Uppsala University, 2019. p. 36
Series
Working paper / Department of Statistics, Uppsala University ; 2019:4
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:uu:diva-382225 (URN)
Note

A revised version was published 2019-09-30.

Available from: 2019-04-23 Created: 2019-04-23 Last updated: 2019-09-30Bibliographically approved

Open Access in DiVA

fulltext(361 kB)61 downloads
File information
File name FULLTEXT01.pdfFile size 361 kBChecksum SHA-512
20ce1b30d9697ca6b5a0aff8c4ea435d4156b6680b63b76aa4dfc99a680662ef1c150c72acf49171fb00d6ac2df58b924d32082ccd3c1c2f383b6ba03806fa4b
Type fulltextMimetype application/pdf
Buy this publication >>

Search in DiVA

By author/editor
Schultzberg, Mårten
By organisation
Department of Statistics
Probability Theory and Statistics

Search outside of DiVA

GoogleGoogle Scholar
Total: 61 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

isbn
urn-nbn

Altmetric score

isbn
urn-nbn
Total: 236 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf