This repository was made as a place to put various projects I have worked on during my graduate studies. A couple were projects I did for classes I took, but but most are projects I worked on outside of my coursework. They demonstrate some of the skills I have developed, but not all.
A colleague of mine was working with some plant data she had collected during her field work. A statistical consultant recommended to her that she use the nonlinear least squares function in R to estimate her model. I wanted to see if the estimates produced by such a model would be biased, and if so, whether or not that bias would be large enough to be of major concern. The relevant files are LogisticNLSBias.pdf and LogisticNLSBias.Rmd
My colleagues at Oregon State University and I created a Data Dashboard for three large climate data sets using Shiny, an R Package for creating interactive web applications. Our webapp took millions of rows of data and distilled them into interactive and informative visualizations. I was responsible for the idea for and execution of the “Decadal Cumulative Precipitation” tool, which helps visualize and understand seasonal and long-term trends in regional precipitation. Web app can be viewed at https://jimmylovestea.shinyapps.io/datadash/
I worked with several of my colleagues at Oregon State University to author a short, high level introduction to the concept of generalized linear models. The tone and language were informal, and the intended audience was students currently enrolled in statistics classes at Oregon State University who are seeking supplementary instruction on the subject. The book can be viewed at https://generalizedregressionmodelingagency.github.io/GRMA/
This piece of code was written as part of a project I did for my class in generalized regression methods in the fall term of 2020. What I wanted to do was demonstrate how Monte-carlo cross validation can be used to estimate the distribution of the accuracy rate (number correct predictions)/(number of total predictions) of a logistic regression model. To do this, I wound up doing the Monte-Carlo cross validation within a holdout cross validation, by setting 100 observations aside to be "new" data that the model would be making predictions on, then performing the Monte-Carlo cross validation on the remaining 599 observations. The purpose of doing it this way was to produce a sort of proof of concept of this type of cross validation to show in my presentation of the project. I used the "biopsy" data set in the MASS package. There were 16 missing values from one variable, so before any cross validation was done I used k-nearest-neighbor hotdeck imputation to impute the missing values.
I am including this rmarkdown file to provide context for the poisson_gifs.Rmd file. This is from a project I worked on in December of 2019 about inhomogenous poisson point processes.
I wrote this code to play around with the gifski package and produce some .gif images. I was motivated to do this by wanting to show how the intensity function of an inhomogenous poisson process could be seen in repeated instances of these processes. The first gif is of a point process in 2 spatial dimensions and 1 temporal dimension. The second image is of several iterations of a 1 dimensional process, with the intensity function plotted above the points.
This is a Shiny app I wrote for a data visualization class I took in the spring of 2020. Baseball has a wide variety of statistics for measuring the performance of the players, but until you're familiar with the game it's difficult to get a sense of what sort of values for these statistics are considered to be "good". So I wrote an application that draws histograms for a given statistic for a given year, where the user has control over the number of histogram bins via a slider. I also made the at-bats a variable that the user has control over, since the data set includes a large number of players who had a small number of at-bats. No one bats 1000, unless they only go to bat once, in which case they only need to hit the ball once to have a batting average of 1.000, which is why I included this widget in the app. The app can be viewed at https://katherine-rose-pulham.shinyapps.io/baseball_app/