My name is Kiana Guarino and I am a 3rd year Quantitative Research Methods doctoral student at Arizona State University. My primary research interests include replication, missing data analysis, study design, and Bayesian methods.
Replication
I am interested in investigating factors than impact replicability (e.g., QRPs, multiplicity) and developing sample size planning methods to increase the likelihood of a successful replication.
Missing Data Analysis
My primary research focus has been on exploring how variations within the broad missing at random (MAR) category of missing data can have differential impacts on statistical power and precision.
Study Design
I aim to develop practical methods for designing adequately powered studies than can account for expected missingness within the randomized controlled trials (RCT) space and beyond.
Bayesian Methods
I am interested in the ways in which we can utilize the Bayesian framework to model uncertainty, incorporate prior knowledge into complex models, and handle missing data.
RESEARCH
Recent Projects
The consequences of optional stopping on the research literature.
ABSTRACT: Optional stopping is the practice of repeatedly analyzing data during the data collection process with the intention of terminating collection once statistical significance is observed. While there are ways to appropriately incorporate interim data analyses, the misuse and lack of transparent reporting of this practice has garnered criticism as a questionable research practice, resulting in inflated Type I error rates. The present simulation study examines optional stopping from the perspective of the overall research literature, where some researchers engage in this practice. Across varying contexts and severities of optional stopping, we examined consequences relevant to replicable research literatures, including effect size bias (partitioned into publication bias and bias unique to optional stopping), heterogeneity across studies, and error rates (including Type S errors). Results demonstrated that optional stopping within a research literature can lead to bias, inaccurate estimates of heterogeneity across studies, and increased Type I and Type S error rates. However, more importantly, patterns emerged that help to demonstrate when and how optional stopping exerts specific consequences on the research literature. We hope these results further clarify the way optional stopping stands in the way of a more cumulative, replicable research literature, and we provide recommendations for researchers in light of this.
Inference in randomized pretest-posttest studies under missing data: Influence of MAR sub-patterns on statistical power and precision.
ABSTRACT: Missing data are common in multiwave studies, including the (two wave) randomized pretest posttest (RPP) design. The commonly assumed missing-at-random (MAR) assumption specifies that the missing data is related only to observed variables. However, limited prior research suggests potential utility in examining missing data patterns within this broad category. This may be particularly important for statistical power and precision: understanding how different forms of MAR affect these quantities could build a foundation for more customized sample size planning. Thus, I investigated how different sub-patterns of the MAR mechanism impact power, precision, bias, and coverage rates when estimating the average treatment effect (ATE) in an RPP design. The present simulation examined how variations in missing data sub-patterns – which link pretest score and/or group membership to dropout at posttest in different ways – affect ATE estimates under listwise deletion and multiple imputation. Results demonstrated that different MAR sub-patterns led to differential impacts on power and precision for both the ATE and within-group regression estimates, although the within-group regression was impacted substantially more. Notably, the patterns associated with the greatest power loss for the ATE did not align with those affecting within-group regression, underscoring the need for more tailored approaches to sample size planning under missing data. These findings support the need for a more nuanced understanding of the MAR assumption, highlighting the importance of developing methods to consider MAR sub-patterns in sample size planning and study design.
Kiana Guarino 