arXiv subjects

Juha Karvanen

Publications and source records attributed to Juha Karvanen.

36 records · Page 2Linked to original sources

arXiv2020

Do-search -- a tool for causal inference and study design with multiple data sources

Epidemiological evidence is based on multiple data sources including clinical trials, cohort studies, surveys, registries and expert opinions. Merging information from different sources opens up new possibilities for the estimation of causal effects. We show how causal effects can be identified and estimated by combining experiments and observations in real and realistic scenarios. As a new tool, we present do-search, a recently developed algorithmic approach that can determine the identifiability of a causal effect. The approach is based on do-calculus, and it can utilize data with non-trivial missing data and selection bias mechanisms. When the effect is identifiable, do-search outputs an identifying formula on which numerical estimation can be based. When the effect is not identifiable, we can use do-search to recognize additional data sources and assumptions that would make the effect identifiable. Throughout the paper, we consider the effect of salt-adding behavior on blood pressure mediated by the salt intake as an example. The identifiability of this effect is resolved in various scenarios with different assumptions on confounding. There are scenarios where the causal effect is identifiable from a chain of experiments but not from survey data, as well as scenarios where the opposite is true. As an illustration, we use survey data from NHANES 2013--2016 and the results from a meta-analysis of randomized controlled trials and estimate the reduction in average systolic blood pressure under an intervention where the use of table salt is discontinued.

stat.AP↗

arXiv2019

Surrogate Outcomes and Transportability

Identification of causal effects is one of the most fundamental tasks of causal inference. We consider an identifiability problem where some experimental and observational data are available but neither data alone is sufficient for the identification of the causal effect of interest. Instead of the outcome of interest, surrogate outcomes are measured in the experiments. This problem is a generalization of identifiability using surrogate experiments and we label it as surrogate outcome identifiability. We show that the concept of transportability provides a sufficient criteria for determining surrogate outcome identifiability for a large class of queries.

cs.AI↗

arXiv2018

Simplifying Probabilistic Expressions in Causal Inference

Obtaining a non-parametric expression for an interventional distribution is one of the most fundamental tasks in causal inference. Such an expression can be obtained for an identifiable causal effect by an algorithm or by manual application of do-calculus. Often we are left with a complicated expression which can lead to biased or inefficient estimates when missing data or measurement errors are involved. We present an automatic simplification algorithm that seeks to eliminate symbolically unnecessary variables from these expressions by taking advantage of the structure of the underlying graphical model. Our method is applicable to all causal effect formulas and is readily available in the R package causaleffect.

stat.ML↗

arXiv2018

Enhancing Identification of Causal Effects by Pruning

Causal models communicate our assumptions about causes and effects in real-world phe- nomena. Often the interest lies in the identification of the effect of an action which means deriving an expression from the observed probability distribution for the interventional distribution resulting from the action. In many cases an identifiability algorithm may return a complicated expression that contains variables that are in fact unnecessary. In practice this can lead to additional computational burden and increased bias or inefficiency of estimates when dealing with measurement error or missing data. We present graphical criteria to detect variables which are redundant in identifying causal effects. We also provide an improved version of a well-known identifiability algorithm that implements these criteria.

stat.ML↗

arXiv2018

Identifying Causal Effects with the R Package causaleffect

Do-calculus is concerned with estimating the interventional distribution of an action from the observed joint probability distribution of the variables in a given causal structure. All identifiable causal effects can be derived using the rules of do-calculus, but the rules themselves do not give any direct indication whether the effect in question is identifiable or not. Shpitser and Pearl constructed an algorithm for identifying joint interventional distributions in causal models, which contain unobserved variables and induce directed acyclic graphs. This algorithm can be seen as a repeated application of the rules of do-calculus and known properties of probabilities, and it ultimately either derives an expression for the causal distribution, or fails to identify the effect, in which case the effect is non-identifiable. In this paper, the R package causaleffect is presented, which provides an implementation of this algorithm. Functionality of causaleffect is also demonstrated through examples.

stat.ME↗

arXiv2017

Adjusting for selective non-participation with re-contact data in the FINRISK 2012 survey

Aims: A common objective of epidemiological surveys is to provide population-level estimates of health indicators. Survey results tend to be biased under selective non-participation. One approach to bias reduction is to collect information about non-participants by contacting them again and asking them to fill in a questionnaire. This information is called re-contact data, and it allows to adjust the estimates for non-participation. Methods: We analyse data from the FINRISK 2012 survey, where re-contact data were collected. We assume that the respondents of the re-contact survey are similar to the remaining non-participants with respect to the health given their available background information. Validity of this assumption is evaluated based on the hospitalization data obtained through record linkage of survey data to the administrative registers. Using this assumption and multiple imputation, we estimate the prevalences of daily smoking and heavy alcohol consumption and compare them to estimates obtained with a commonly used assumption that the participants represent the entire target group. Results: This approach produces higher prevalence estimates than what is estimated from participants only. Among men, smoking prevalence estimate was 28.5% (23.2% for participants), heavy alcohol consumption prevalence was 9.4% (6.8% for participants). Among women, smoking prevalence was 19.0% (16.5% for participants) and heavy alcohol consumption 4.8% (3.0% for participants). Conclusion: Utilization of re-contact data is a useful method to adjust for non-participation bias on population estimates in epidemiological surveys.

stat.AP↗

arXiv2017

Optimal design of observational studies: overview and synthesis

We review typical design problems encountered in the planning of observational studies and propose a unifying framework that allows us to use the same concepts and notation for different problems. In the framework, the design is defined as a probability measure in the space of observational processes that determine whether the value of a variable is observed for a specific unit at the given time. The optimal design is then defined, according to Bayesian decision theory, to be the one that maximizes the expected utility related to the design. We present examples on the use of the framework and discuss methods for deriving optimal or approximately optimal designs.

math.ST↗

arXiv2017

Bayesian models for data missing not at random in health examination surveys

In epidemiological surveys, data missing not at random (MNAR) due to survey nonresponse may potentially lead to a bias in the risk factor estimates. We propose an approach based on Bayesian data augmentation and survival modelling to reduce the nonresponse bias. The approach requires additional information based on follow-up data. We present a case study of smoking prevalence using FINRISK data collected between 1972 and 2007 with a follow-up to the end of 2012 and compare it to other commonly applied missing at random (MAR) imputation approaches. A simulation experiment is carried out to study the validity of the approaches. Our approach appears to reduce the nonresponse bias substantially, where as MAR imputation was not successful in bias reduction.

stat.ME↗

arXiv2016

Bayesian subcohort selection for longitudinal covariate measurements in follow-up studies

We consider planning longitudinal covariate measurements in follow-up studies where covariates are time-varying. We assume that the entire cohort cannot be selected for longitudinal measurements due to financial limitations and study how a subset of the cohort should be selected optimally in order to obtain precise estimates of covariate effects in a survival model. In our approach, the study will be designed sequentially utilizing the data collected in previous measurements of the individuals as prior information. We propose using a Bayesian optimality criterion in the subcohort selections, which is compared with simple random sampling using simulated and real follow-up data. This study extends previous results where optimal subcohort selection was studied with only one re-measurement and one covariate, to more realistic cases where several covariates and measurement points are allowed. Our results support the conclusion that the precision of the estimates can be clearly improved by optimal design.

stat.ME↗

arXiv2016

Prioritizing covariates in the planning of future studies in the meta-analytic framework

Science can be seen as a sequential process where each new study augments evidence to the existing knowledge. To have the best prospects to make an impact in this process, a new study should be designed optimally taking into account the previous studies and other prior information. We propose a formal approach for the covariate prioritization, i.e., the decision about the covariates to be measured in a new study. The decision criteria can be based on conditional power, change of the p-value, change in lower confidence limit, Kullback-Leibler divergence, Bayes factors, Bayesian false discovery rate or difference between prior and posterior expectation. The criteria can be also used for decisions on the sample size. As an illustration, we consider covariate prioritization based on genome-wide association studies for C-reactive protein levels and make suggestions on the genes to be studied further. keywords: design; evidence-based medicine; meta-analysis; power; scientific method

stat.ME↗

arXiv2015

Correcting for non-ignorable missingness in smoking trends

Data missing not at random (MNAR) is a major challenge in survey sampling. We propose an approach based on registry data to deal with non-ignorable missingness in health examination surveys. The approach relies on follow-up data available from administrative registers several years after the survey. For illustration we use data on smoking prevalence in Finnish National FINRISK study conducted in 1972-1997. The data consist of measured survey information including missingness indicators, register-based background information and register-based time-to-disease survival data. The parameters of missingness mechanism are estimable with these data although the original survey data are MNAR. The underlying data generation process is modelled by a Bayesian model. The results indicate that the estimated smoking prevalence rates in Finland may be significantly affected by missing data.

stat.AP↗

arXiv2014

Estimating complex causal effects from incomplete observational data

Despite the major advances taken in causal modeling, causality is still an unfamiliar topic for many statisticians. In this paper, it is demonstrated from the beginning to the end how causal effects can be estimated from observational data assuming that the causal structure is known. To make the problem more challenging, the causal effects are highly nonlinear and the data are missing at random. The tools used in the estimation include causal models with design, causal calculus, multiple imputation and generalized additive models. The main message is that a trained statistician can estimate causal effects by judiciously combining existing tools.

stat.ME↗

arXiv2014

Survey data and Bayesian analysis: a cost-efficient way to estimate customer equity

We present a Bayesian framework for estimating the customer lifetime value (CLV) and the customer equity (CE) based on the purchasing behavior deducible from the market surveys on customer purchasing behavior. The proposed framework systematically addresses the challenges faced when the future value of customers is estimated based on survey data. The scarcity of the survey data and the sampling variance are countered by utilizing the prior information and quantifying the uncertainty of the CE and CLV estimates by posterior distributions. Furthermore, information on the purchase behavior of the customers of competitors available in the survey data is integrated to the framework. The introduced approach is directly applicable in the domains where a customer relationship can be thought to be monogamous. As an example on the use of the framework, we analyze a consumer survey on mobile phones carried out in Finland in February 2013. The survey data contains consumer given information on the current and previous brand of the phone and the times of the last two purchases.

stat.AP↗

arXiv2014

Study design in causal models

The causal assumptions, the study design and the data are the elements required for scientific inference in empirical research. The research is adequately communicated only if all of these elements and their relations are described precisely. Causal models with design describe the study design and the missing data mechanism together with the causal structure and allow the direct application of causal calculus in the estimation of the causal effects. The flow of the study is visualized by ordering the nodes of the causal diagram in two dimensions by their causal order and the time of the observation. Conclusions whether a causal or observational relationship can be estimated from the collected incomplete data can be made directly from the graph. Causal models with design offer a systematic and unifying view scientific inference and increase the clarity and speed of communication. Examples on the causal models for a case-control study, a nested case-control study, a clinical trial and a two-stage case-cohort study are presented.

stat.ME↗

arXiv2008

Approximate cost-efficient sequential designs for binary response models with application to switching measurements

The efficiency of an experimental design is ultimately measured in terms of time and resources needed for the experiment. Optimal sequential (multi-stage) design is studied in the situation where each stage involves a fixed cost. The problem is motivated by switching measurements on superconducting Josephson junctions. In this quantum mechanical experiment, the sequences of current pulses are applied to the Josephson junction sample and a binary response indicating the presence or the absence of a voltage response is measured. The binary response can be modeled by a generalized linear model with the complementary log-log link function. The other models considered are the logit model and the probit model. For these three models, the approximately optimal sample size for the next stage as a function of the current Fisher information and the stage cost is determined. The cost-efficiency of the proposed design is demonstrated in simulations based on real data from switching measurements. The results can be directly applied to switching measurements and they may lead to substantial savings in the time needed for the experiment.

math.ST↗

arXiv2008

Efficient initial designs for binary response data

In this paper we introduce a binary search algorithm that efficiently finds initial maximum likelihood estimates for sequential experiments where a binary response is modeled by a continuous factor. The problem is motivated by switching measurements on superconducting Josephson junctions. In this quantum mechanical experiment, the current is the factor controlled by the experimenter and a binary response indicating the presence or the absence of a voltage response is measured. The prior knowledge on the model parameters is typically poor, which may cause the common approaches of initial estimation to fail. The binary search algorithm is designed to work reliably even when the prior information is very poor. The properties of the algorithm are studied in simulations and an advantage over the initial estimation with equally spaced factor levels is demonstrated. We also study the cost-efficiency of the binary search algorithm and find the approximately optimal number of measurements per stage when there is a cost related to the number of stages in the experiment. KEY WORDS: optimal design, binary search, logistic regression, complementary log-log, quantum physics, switching measurement

math.ST↗

arXiv2007

Characterizing the generalized lambda distribution by L-moments

The generalized lambda distribution (GLD) is a flexible four parameter distribution with many practical applications. L-moments of the GLD can be expressed in closed form and are good alternatives for the central moments. The L-moments of the GLD up to an arbitrary order are presented, and a study of L-skewness and L-kurtosis that can be achieved by the GLD is provided. The boundaries of L-skewness and L-kurtosis are derived analytically for the symmetric GLD and calculated numerically for the GLD in general. Additionally, the contours of L-skewness and L-kurtosis are presented as functions of the GLD parameters. It is found that with an exception of the smallest values of L-kurtosis, the GLD covers all possible pairs of L-skewness and L-kurtosis and often there are two or more distributions that share the same L-skewness and the same L-kurtosis. Examples that demonstrate situations where there are four GLD members with the same L-skewness and the same L-kurtosis are presented. The estimation of the GLD parameters is studied in a simulation example where method of L-moments compares favorably to more complicated estimation methods. The results increase the knowledge on the distributions that belong to the GLD family and can be utilized in model selection and estimation.

math.ST↗

arXiv2006

Experimental Designs for Binary Data in Switching Measurements on Superconducting Josephson Junctions

We study the optimal design of switching measurements of small Josephson junction circuits which operate in the macroscopic quantum tunnelling regime. Starting from the D-optimality criterion we derive the optimal design for the estimation of the unknown parameters of the underlying Gumbel type distribution. As a practical method for the measurements, we propose a sequential design that combines heuristic search for initial estimates and maximum likelihood estimation. The presented design has immediate applications in the area of superconducting electronics implying faster data acquisition. The presented experimental results confirm the usefulness of the method. KEY WORDS: optimal design, D-optimality, logistic regression, complementary log-log link, quantum physics, escape measurements

cond-mat.supr-con↗