Aleksandra Kazakova

The following post is part of our Student Spotlight Initiative

Aleksandra Kazakova is a Phd Candidate in the Quantitative Methods in Educational and Psychological Research Specialization. Recently, Aleksandra presented a poster at the 2021 Conference on Statistical Practice titled, “An application of Structural Equation Modeling in the analysis of ordered categorical factors indicating the relationship between mother-child, father-child, and mother-father interactions impacting cognitive development of US children at age five.” 

This study was designed to reexamine the influence of parental involvement in a child’s life on the child’s receptive speech, revealing the moderators’ effects along with the direct impact of father-child, mother-child, and father-mother interactions by utilizing the Structural Equation Modeling. An application of SEM allowed us to model unexplained variances and to analyze multiple potential factors of children’s vocabulary simultaneously, linking micro and macro-perspectives to avoid misleading interpretations of the factors’ effects. While classic estimation methods in SEM assume continuous normally distributed variables, the data collected for this research is mostly ordinal in nature. So, the design of this study specifically accounted for the ordinal variables used. The results revealed that when investigating the ‘fathers and mother’s degree of involvement predictors of child’s vocabulary, the ‘status of mother and father cohabitation’, ‘mother’s age’, and ‘father’s attitude towards mother’ variables should not be omitted in the model. Otherwise, the results could be misleading, inflating the effect of the father/mother’s involvement in a child’s activities on a child’s speech. 

Here is a link to her poster:

Here is what Aleksandra is doing now:

“I am currently working on a Multiverse analysis of the same factors to continue exploring the issue of biased results that are not robust and impossible to replicate. My goal is to investigate the problem of choosing the right analytical approach. Even after the study design is determined and the data is collected, there are still various ways to conduct the analysis based on the combination of computational tools, models, variables selection, statistical assumptions, and choices made towards the missing data imputation. Thus, to avoid misleading results, I proceeded with the Multiverse approach while expanding the initially selected sample to two time-levels: children at age five and nine. At the end of my research, the Multiverse analysis’s overall effect will be measured and juxtaposed to the results from every single statistical approach independently performed. Such comparison will allow checking the assumption that each choice made towards the study method selection could drastically skew the outcome by inflating some factors’ importance and downsizing the others’ effect.” 

Comment below if you have questions or comments about Aleksandra’s research! 

This entry is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license.