Variance Estimation Research Articles

Estimating the Complier Average Causal Effect with Non-Ignorable Missing Outcomes Using Likelihood Analysis

Missing data problems arise in randomized trials, which complicates the inference of causal effects if the missing mechanism is non-ignorable. We tackle the challenge of identifying and estimating the complier average causal effect parameters under non-ignorable missingness by increasing the covariates to mitigate the sensitivity to the violation of specific identification assumptions. The missing data mechanism is assumed to follow a logistic model, wherein the absence of the outcome is explained by the outcome itself, the treatment received, and the covariates. We establish the identifiability of the models under mild conditions by assuming that the outcome follows a normal distribution. We develop a computational method to estimate model parameters through a two-step likelihood estimation approach, employing subgroup analysis. The bootstrap method is employed for variance estimation, and the effectiveness of our approach is confirmed through simulation. We applied the proposed method to analyze the household income dataset from the Chinese Household Income Project Survey 2013.

The performance of marginal structural models for estimating risk differences and relative risks using weighted univariate generalized linear models.

We used Monte Carlo simulations to compare the performance of marginal structural models (MSMs) based on weighted univariate generalized linear models (GLMs) to estimate risk differences and relative risks for binary outcomes in observational studies. We considered four different sets of weights based on the propensity score: inverse probability of treatment weights with the average treatment effect as the target estimand, weights for estimating the average treatment effect in the treated, matching weights and overlap weights. We considered sample sizes ranging from 500 to 10,000 and allowed the prevalence of treatment to range from 0.1 to 0.9. We examined both the robust variance estimator when using generalized estimating equations with an independent working correlation matrix and a bootstrap variance estimator for estimating the standard error of the risk difference and the log-relative risk. The performance of these methods was compared with that of direct weighting. Both the direct weighting approach and MSMs based on weighted univariate GLMs resulted in the identical estimates of risk differences and relative risks. When sample sizes were small to moderate, the use of an MSM with a bootstrap variance estimator tended to result in the most accurate estimates of standard errors. When sample sizes were large, the direct weighting approach and an MSM with a bootstrap variance estimator tended to produce estimates of standard error with similar accuracy. When using a MSM to estimate risk differences and relative risks, in general it is preferable to use a bootstrap variance estimator than the robust variance estimator. We illustrate the application of the different methods for estimating risks differences and relative risks using an observational study on the effect on mortality of discharge prescribing of a beta-blocker in patients hospitalized with acute myocardial infarction.

Heavy alcohol use and the HIV care continuum in Kenya: a population-based study

ABSTRACT Heavy alcohol use (HAU) can destabilize engagement along the HIV care continuum. Population-based studies assessing associations of HAU with HIV treatment outcomes are lacking, especially in sub-Saharan Africa. We leveraged data from the Kenya Population-based HIV Impact Assessment to identify associations of self-reported HAU, assessed using two items measuring the frequency and quantity of past-year alcohol consumption, with serum biomarkers for HIV serostatus unawareness, antiretroviral therapy (ART) non-use, and HIV viremia (≥1000 RNA copies/mL). Overall and sex-stratified survey-weighted logistic regression with jackknife variance estimation modeled adjusted odds ratios (adjOR) of HIV treatment indicators by HAU. Overall, 1491 persons living with HIV aged 15–64 years (68.4% female) were included. The prevalence of HAU was 8.9% (95% confidence interval [95%CI]: 6.8–11.0%) and was significantly more pronounced in males than females (19.6% vs. 4.0%, p < 0.001). In multivariable analysis, HAU was significantly (p < 0.001) associated with HIV serostatus unawareness (adjOR = 3.65, 95%CI: 2.14–6.23), ART non-use (adjOR = 3.81, 95%CI: 2.25–6.43), and HIV viremia (adjOR = 3.13, 95%CI: 1.85–5.32). Incorporating sex-specific alcohol use screening into HIV testing and treatment services in populations where HAU is prevalent could optimize clinical outcomes along the HIV care continuum.

Statistical Methods for Pooling ages of End-Stage-Renal-Disease patients The Examples of India and Malaysia

In pharmacoeconomic literature, age has been shown to be an important variate concerning cost analyses of End-Stage-Renal-Disease (ESRD) patients. In the present article, pooling of data relative to ages of ESRD patients is proposed as a method for estimating typical age parameters like means and variances, and also for comparing age differences between countries. Statistical techniques for mean and variance estimation, large sample statistical theory, confidence intervals for means, and parametric tests for statistical inference concerning comparison between means are used, and the main advantages of pooling are investigated. Homogeneity of the pooled data is also discussed using mixture models. As an example of application, data obtained from four countries were included into our analysis. Results showed that pooling of data increases the power of the tests used for statistical inference, apart from providing a better accuracy for the estimates of the means. Thus, statistical results are noticeably improved when pooling of data is used.

The effectiveness of abstinence-based and harm reduction-based interventions in reducing problematic substance use in adults who are experiencing homelessness in high income countries: A systematic review and meta-analysis: A systematic review.

The effectiveness of abstinence-based and harm reduction-based interventions in reducing problematic substance use in adults who are experiencing homelessness in high income countries: A systematic review and meta-analysis: A systematic review.

Analyze as randomized—Why dropping block effects in designed experiments is a bad idea

AbstractAgricultural experiments are often laid out as blocked designs such as the randomized complete block designs (RCBD) or split‐plot designs (SPD). Statistical analysis should follow the principle “analyze as randomized.” However, block effects are often not modeled or are typically dropped from the model when non‐significant. Additionally, if linear mixed models are fitted by REML with a non‐negativity constraint on all variances, an implicit model reduction occurs when the variance estimate for random block effects becomes negative in the final iteration. This study investigates the consequences of these types of model reduction on the Type I error rate and the standard error of a treatment difference (s.e.d.) by Monte Carlo simulation for experiments designed as RCBD or SPD. The number of blocks and treatments and the ratio of block or main‐plot error variance to residual error variance were varied, resulting in 27 scenarios. Dropping the block effect by default resulted in deflated Type I error rates and increased the s.e.d. in the RCBD. For the SPD, Type I error rates were inflated for the sub‐plot factor and the main‐plot‐by‐sub‐plot factor interaction, but were deflated for the main‐plot factor. Adverse effects of model reduction were reduced, but did not vanish completely, for model reduction based on significance testing. Implicit model reduction led to inflated Type I error rates for small variance ratios and small datasets. We conclude that block effects as well as main‐plot error effects need to be included by all means in statistical analysis.

Patients with Inflammatory Bowel Disease who regularly consume fruits and vegetables present lower prevalence of disease activation: A cross-sectional study

BackgroundExclusion diets are common practices among individuals with Inflammatory Bowel Disease (IBD). Reports that certain foods trigger or worsen symptoms are recurrent but lack evidence. The aim of the study was to identify which foods were most frequently avoided by patients with Crohn's Disease (CD) and Ulcerative Colitis (UC) and whether the consumption of any food group was associated with disease activity. MethodsCross-sectional study with adult patients seen at an outpatient clinic in a tertiary public hospital. Dietary intake and eating habits were accessed through questionnaires administered via telephone interview. Disease activity and symptoms were assessed using the Harvey-Bradshaw Index (IHB) for CD and the Lichtiger Index (LI) for UC. Poisson regression with a robust variance estimator was used to estimate prevalence ratios. Analyzes were performed using SPSS - Statistical Package for the Social Sciences. ResultsThe study included 145 patients. Of these, 69.7% avoided certain foods, with citrus fruits and raw vegetables among the most avoided (16.8% and 13.8%, respectively). Regular consumption of fruits (PR= 0.56; CI 95% 0.32 – 0.97; p= 0.042) and vegetables (PR= 0.56; CI 95% 0.32 – 0.98; p= 0.045) was associated with a 44% lower prevalence of the active phase of the disease, compared to those who do not consume these foods, adjusted for age, sex and type of disease. Other food items did not present significant associations in the adjusted model. ConclusionsFruit and vegetable intake appears to have a protective role in the recurrence of IBD. Excluding foods is a common practice, even among patients in remission, and this should be combated as it can lead to nutritional losses. It is important to reinforce with patients the benefits of a varied and less restrictive diet.

Computationally Efficient Demographic History Inference from Allele Frequencies with Supervised Machine Learning.

Inferring past demographic history of natural populations from genomic data is of central concern in many studies across research fields. Previously, our group had developed dadi, a widely used demographic history inference method based on the allele frequency spectrum (AFS) and maximum composite-likelihood optimization. However, dadi's optimization procedure can be computationally expensive. Here, we present donni (demography optimization via neural network inference), a new inference method based on dadi that is more efficient while maintaining comparable inference accuracy. For each dadi-supported demographic model, donni simulates the expected AFS for a range of model parameters then trains a set of Mean Variance Estimation neural networks using the simulated AFS. Trained networks can then be used to instantaneously infer the model parameters from future genomic data summarized by an AFS. We demonstrate that for many demographic models, donni can infer some parameters, such as population size changes, very well and other parameters, such as migration rates and times of demographic events, fairly well. Importantly, donni provides both parameter and confidence interval estimates from input AFS with accuracy comparable to parameters inferred by dadi's likelihood optimization while bypassing its long and computationally intensive evaluation process. donni's performance demonstrates that supervised machine learning algorithms may be a promising avenue for developing more sustainable and computationally efficient demographic history inference methods.

Serum biomarker levels in smokers and non-smokers following periodontal therapy. A prospective cohort study

BackgroundTo compare presence and levels of serum cytokines in smokers and non-smokers with periodontitis following periodontal therapy.MethodsThirty heavy smokers and 30 non-smokers with stage III or IV periodontitis were included in this prospective cohort study. Clinical data and blood serum were collected at baseline (T0), after step I-III (T1), and after 12 months step IV periodontal therapy (T2). Cytokine IL-1β, IL-6, IL-8, TNF-α, IL-10, and IP-10 levels were measured using multiplex kit Bio-Plex Human Pro™ Assay. Linear regression models with cluster robust variance estimates to adjust for repeated observations were used to test intra- and intergroup levels for each marker, IL-6 and IL-8 defined as primary outcomes.ResultsClinical outcomes improved in both groups following therapy (p < 0.05). IL-6 levels increased with 75.0% from T0-T2 among smokers (p = 0.004). No significant intra- or intergroup differences were observed for IL-8. Higher levels of TNF-α (44.1%) and IL-10 (50.6%) were detected in smokers compared with non-smokers at T1 (p = 0.007 and p = 0.037, respectively). From T1-T2, differences in mean change over time for levels of TNF-α and IL-10 were observed in smokers compared with non-smokers (p = 0.005 and p = 0.008, respectively).ConclusionUpregulated levels of serum cytokines in smokers indicate a systemic effect of smoking following periodontal therapy. Differences in cytokine levels between smokers and non-smokers demonstrate a smoking induced modulation of specific systemic immunological responses in patients with severe periodontitis.

Characterising X-ray variability in light curves with complex sampling patterns: Application to the eROSITA south ecliptic pole survey

Aims : During its all-sky survey phase, the extended ROentgen Survey with an Imaging Telescope Array (eROSITA) X-ray telescope on board the Spectrum-Roentgen-Gamma (SRG) spacecraft scans through the ecliptic poles every 4 hours. This extensive data set of long-duration, frequent, and consistent observations of thousands of X-ray sources is ideal for a detailed long-term X-ray-variability analysis. However, individual observations are short, are separated by long but consistent gaps, and have varying exposure times. Therefore, the identification of variable sources and the characterisation and quantification of their variability requires a unique methodology. We aim to develop and evaluate variability analysis methods for eROSITA observations, focusing on sources close to the survey poles. We also aim to detect intrinsically variable sources at any count rate and quantify the variability of low-count-rate sources. Methods : We simulate eROSITA-like light curves to evaluate and quantify the effect of survey mode observations on the measured periodogram and normalised excess variance. We introduce a new method for estimating the normalised intrinsic variance of a source based on the Bayesian excess variance (bexvar) method. Results : We determine thresholds for identifying likely variable sources while minimising the false-positive rate, as a function of the number of bins, and the average count rate in the light curve. The bexvar normalised intrinsic variance estimate is significantly more accurate than the normalised excess variance method in the Poisson regime. At high count rates, the two methods are comparable. We quantify the scatter in the intrinsic variance of a stationary pink-noise process, and investigate how to reduce it. Finally, we determine a description of the excess noise in a periodogram caused by varying exposure times throughout a light curve. Although most of these methods were developed specifically for analysing variable active galactic nuclei in the eROSITA all-sky survey, they can also be used for the variability analysis of other datasets from other telescopes, with slight modifications.

On variance estimation of the inverse probability-of-treatment weighting estimator: A tutorial for different types of propensity score weights.

Propensity score methods, such as inverse probability-of-treatment weighting (IPTW), have been increasingly used for covariate balancing in both observational studies and randomized trials, allowing the control of both systematic and chance imbalances. Approaches using IPTW are based on two steps: (i) estimation of the individual propensity scores (PS), and (ii) estimation of the treatment effect by applying PS weights. Thus, a variance estimator that accounts for both steps is crucial for correct inference. Using a variance estimator which ignores the first step leads to overestimated variance when the estimand is the average treatment effect (ATE), and to under or overestimated estimates when targeting the average treatment effect on the treated (ATT). In this article, we emphasize the importance of using an IPTW variance estimator that correctly considers the uncertainty in PS estimation. We present a comprehensive tutorial to obtain unbiased variance estimates, by proposing and applying a unifying formula for different types of PS weights (ATE, ATT, matching and overlap weights). This can be derived either via the linearization approach or M-estimation. Extensive R code is provided along with the corresponding large-sample theory. We perform simulation studies to illustrate the behavior of the estimators under different treatment and outcome prevalences and demonstrate appropriate behavior of the analytical variance estimator. We also use a reproducible analysis of observational lung cancer data as an illustrative example, estimating the effect of receiving a PET-CT scan on the receipt of surgery.

The effect of process variability and data quality on performance of a state-space stock assessment model

State-space modeling is an emerging approach to age structured fisheries stock assessment that can accommodate multiple sources of variability in processes like recruitment, abundance, and selectivity. By maximizing the marginal likelihood by treating yearly deviations as random effects and then integrating them from the likelihood, these models can estimate multiple process variances. Several fisheries software packages have been developed that use a state-space framework with marginal likelihood, which has increased their popularity and usage across the U.S. Atlantic coast, Canada, and Europe. However, robust testing is still needed to gauge the applicability of these models and understand how they perform under a range of realistic variability in the process or observation error. Using an assessment model fit to Gulf of Maine Haddock as a baseline, we used a simulation-estimation procedure to test if state-space stock assessment models could produce approximately unbiased and precise estimates over a range of process variances that extended from zero to well above the levels estimated in the Gulf of Maine Haddock assessment, or when observations were noisier (i.e., more variable around their true value) than had been assumed in the assessment. We fit alternative estimation models that differed in which processes errors were included (of recruitment, expected survival, and fishery selectivity). State-space models which specify random effects in all three processes produced approximately unbiased and precise estimates of biomass and exploitation for most operating models and therefore are recommended except when variability in expected survival is absent (or very low), in which case the model is unlikely to converge. A conventional statistical-catch-at-age model with recruitment estimated as a fixed effect for each year, deterministic expected survival, and constant selectivity produced estimates that were comparable to the best state-space model, but do not provide internal estimates of process variances and did not perform well when recruitment was highly variable. This work will facilitate the use of state-space stock assessment models and choosing the parameterization that will produce the most accurate output to inform future predictions and management.

The Public Purposes of Private Education: a Civic Outcomes Meta-Analysis

Since Plato and Aristotle, political theorists have discussed the important role of education in forming democratic citizens. They disagree, however, over whether public or private schools are more effective at nurturing citizenship. We conduct a statistical meta-analysis to identify the average association between private schooling and measures of four central civic outcomes: political tolerance, political participation, civic knowledge and skills, and voluntarism and social capital. Our search identifies 13,301 initial target studies, ultimately yielding 531 effects from 57 qualified studies drawing from 40 different databases. Using Robust Variance Estimation, we determine that, on average, private schooling boosts any civic outcome by 0.055 standard deviations over public schooling. Religious private schooling, particularly, is strongly associated with positive civic outcomes. The evidence is especially strong that private schooling is correlated with higher levels of political tolerance and political knowledge and skills. We discuss heterogeneities, robustness checks, and implications.

Evaluation of non-additive genetic effects on carcass and meat quality traits in Korean Hanwoo cattle using genomic models

The traditional genetic evaluation methods generally consider additive genetic effects only and often ignore non-additive (dominance and epistasis) effects that may have contributed to genetic variation of complex traits of livestock species. The available dense single nucleotide polymorphisms (SNPs) panels offer to investigate the potential benefits of including non-additive genetic effects in the genomic evaluation models. Data from 16 971 genotyped (Illumina Bovine 50 K SNP chip) Korean Hanwoo cattle were used to estimate genetic variance components and prediction accuracy of genomic breeding values (GEBVs) for four carcass and meat quality traits: carcass weight (CWT), eye muscle area (EMA), back fat thickness (BFT) and marbling score (MS). Five different genetic models were evaluated through including additive, dominance and epistatic interactions (additive by additive, A × A; additive by dominance, A × D and dominance by dominance, D × D) successively in the models. The estimates of additive genetic variances and narrow sense heritabilities (ha2) were found similar across the evaluated models and traits except when additive interaction (A × A) was included. The dominance variance estimates relative to phenotypic variance ranged from 1.7–3.4% for CWT and MS traits, whereas, they were close to zero for EMA and BFT traits. The magnitude of A × A epistatic heritability (haa2) ranged between 14.8 and 27.7% in all traits. However, heritability estimates for A × D and D × D epistatic interactions (had2 and hdd2) were quite low compared to haa2 and were contributed only 0.0–9.7% of the total phenotypic variation. In general, broad sense heritability (hG2) estimates were almost twice (ranging between 0.54 and 0.68) the ha2 for all of the investigated traits. The inclusion of dominance effects did not improve the prediction accuracy of GEBV but improved 2.0–3.0% when epistatic effects were included in the model. More importantly, rank correlation revealed that partitioning of variance components considering dominance and epistatic effects in the model would enable to re-rank of top animals with better prediction of GEBV. The present result suggests that dominance and epistatic effects could be included in the genomic evaluation model for better estimates of variance components and more accurate prediction of GEBV for carcass and meat quality traits in Korean Hanwoo cattle.

The effect of game-based approaches on decision-making, knowledge, and motor skill: A systematic review and a multilevel meta-analysis

This study aimed to compare the effect between game-based approaches (GBAs) and traditional skill approaches on decision-making, knowledge and motor skill in physical education students and athletes. A systematic review and meta-analysis of experimental studies available before October 2023 was conducted. The initial search yielded 8431 articles, with 28 articles ( n = 1600) meeting the inclusion criteria. Studies were analyzed using three-level random-effects models with a robust variance estimation. Outcomes were computed as raw mean differences and Hedges’s g effect sizes. Results indicate that GBAs have a positive heterogeneous effect on decision-making in game situations (ES = 11.41%; 95% CI [4.39, 18.43]) and motor skill in skill tests ( g = 0.36; 95% CI [0.14, 0.57]). GBAs did not have an effect on knowledge ( g = 0 .37; 95% CI [−0.12, 0.86]) or motor skill in game situations (ES = 1.13%; 95% CI [−2.43, 4.68]). Meta-regression analyses revealed that the experience of the interventionist, the quality of the studies, and the comparison condition significantly influence the impact of GBAs on motor skill tests. More detailed and transparent reporting of trials would benefit the field.

Information-based group sequential design for post-market safety monitoring of medical products using real world data.

Real world healthcare data are commonly used in post-market safety monitoring studies to address potential safety issues related to newly approved medical products. Such studies typically involve repeated evaluations of accumulating safety data with respect to pre-defined hypotheses, for which the group sequential design provides a rigorous and flexible statistical framework. A major challenge in designing a group sequential safety monitoring study is the uncertainty associated with product uptake, which makes it difficult to specify the final sample size or maximum duration of the study. To deal with this challenge, we propose an information-based group sequential design which specifies a target amount of information that would produce adequate power for detecting a clinically significant effect size. At each interim analysis, the variance estimate for the treatment effect of interest is used to compute the current information time, and a pre-specified alpha spending function is used to determine the stopping boundary. The proposed design can be applied to regression models that adjust for potential confounders and/or heterogeneous treatment exposure. Simulation results demonstrate that the proposed design performs reasonably well in realistic settings.

Using Extant Data to Improve Estimation of the Standardized Mean Difference

This article presents methods for using extant data to improve the properties of estimators of the standardized mean difference (SMD) effect size. Because samples recruited into education research studies are often more homogeneous than the populations of policy interest, the variation in educational outcomes can be smaller in these samples than is reflective of the true variation in the population. This affects effect size estimation since the sample standard deviation is used in the denominator of the SMD. We propose leveraging extant data on sample variance estimates from multiple studies, made available via clearinghouse databases such as the What Works Clearinghouse, to standardize a mean difference. This allows effect sizes to be benchmarked across a common and broad population, thus enabling better comparability across studies and interventions. We derive the new estimators of the population variance and the corresponding SMD, which pool sample variances from multiple studies using both an analysis of variance and a meta-analytic framework. We demonstrate the properties of these estimators via analytic and simulation results and offer recommendations for when these estimators are appropriate in practice.

Variance estimation for average treatment effects estimated by g-computation

AbstractThe average treatment effect is used to evaluate effects of interventions in a population. Under certain causal assumptions, such an effect may be estimated from observational data using the g-computation technique. The asymptotic properties of this estimator appears not to be well-known and hence bootstrapping has become the preferred method for estimating its variance. Bootstrapping is, however, not an optimal choice for multiple reasons; it is a slow procedure and, if based on too few bootstrap samples, results in a highly variable estimator of the variance. In this paper, we consider estimators of potential outcome means and average treatment effects using g-computation. We consider these parameters for the entire population but also in subgroups, for example, the average treatment effect among the treated. We derive their asymptotic distributions in a general framework. An estimator of the asymptotic variance is proposed and shown to be consistent when g-computation is used in conjunction with the M-estimation technique. The proposed estimator is shown to be superior to the bootstrap technique in a simulation study. Robustness against model misspecification is also demonstrated by means of simulations.

Improvement in variance estimation using transformed auxiliary variable under simple random sampling

This paper offers a novel approach to formulate efficient ratio estimator of the population variance using a transformed auxiliary variable. The impact of transformation on auxiliary information has also been discussed. It is observed that incorporating a transformed auxiliary variable result in a high gain in efficiency. Theoretical properties of the newly developed estimators have been derived. The empirical and simulation studies show that the suggested estimators outperformed the existing estimators.

Can increasing preventive patrol in large geographic areas reduce crime? A systematic review and meta‐analysis

AbstractResearch summaryWe conducted a systematic review and meta‐analysis examining whether increasing preventive patrol in large areas reduces crime. Our review included experimental and quasi‐experimental studies that focused on areas such as beats, precincts, or entire jurisdictions and that measured a crime outcome either through official data or surveys. We identified 17 studies to include in our review. We used two methods for assessing study impacts: an approach which identified a primary/general outcome measure and a second approach which used robust variance estimation (RVE) and included all effect sizes across each study. Both approaches showed small crime prevention benefits (RVE: 9% decline; primary/general: 6% decline), but only the RVE model was significant at conventional levels (p &lt; 0.05). There was no significant evidence of displacement. Moderator analyses suggest that as dosage increases so do the crime prevention impacts. In RVE models, preventive patrol was associated with significant reductions in property and violent crime, but nonsignificant increases in drug and disorder offenses.Policy implicationsIncreasing preventive patrol activities has the potential to reduce crime in large administrative areas. At the same time, existing studies offer little guidance as to how such preventive patrol should be carried out. Deterrence theory, as well as evidence from studies of hot spots policing, suggests that the greatest benefits will be gained from informing patrol efforts about where and when crime occurs. Although more research is needed regarding patrol allocations in large areas, present knowledge suggests that the more such patrols can be targeted at specific places at specific times, the greater will be the crime control benefits. In this context, we argue that police agencies may want to apply a hybrid approach to police patrol, which would include a combination of hot spots policing units and general patrol units informed by data on where crime is concentrated.