In dosc91/SfL: Statistics for Linguistics

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(SfL)
library(DT)
library(lme4)

The SfL package was created to accompany the Statistics for Linguistics online workshop in August/September 2021. The workshop was a satellite event to the 3rd Forensic Linguistics Short Course.

This vignette gives an overview of functions included in the SfL package. Please refer to the data set vignette for an overview of included data sets.

---

Overview

This is a full list of all functions currently contained in SfL:

• correlation_matrix
• create_error_bar_df
• open_exercise
• open_slides
• predictor_competition
• predictor_strength
• tukey

Where necessary, the data_s data set will be used for illustrations in this vignette.

data("data_s")

---

Create Correlation Matrix {#correlation_matrix}

The correlation_matrix function creates a matrix of scatter plots with Pearson and Spearman correlations in the lower triangle. It is a wrapper for languageR::pairscor.fnc.

The function takes a data set and a vector of column (variable) names to display in the plot.

correlation_matrix(data = data_s, variables = c("typeOfS", "pauseBin", "sDur"))

---

Create Error Bar Dataframe {#create_error_bar_df}

The create_error_bar_df function creates a dataframe with which a ggplot2 bar plot (geom_bar) with error bars (+/- 1 standard deviation) can easily be created.

The function takes an original data set, a numerical value to summarise, and one or more categorical variables as arguments.
Additionally, the user may specify whether they want to compute the standard deviation (default) or standard error, and whether they wish to compute the base standard deviation/error (default) or a multiple of that, e.g. two times the standard deviation.

# example 1: one categorical variable

df1 <- create_error_bar_df(data = data_s, numerical = "sDur", factors = "pauseBin")

df1

# example 2: more than one categorical variable

df2 <- create_error_bar_df(data = data_s, numerical = "sDur", factors = c("pauseBin", "typeOfS", "folType"))

df2

# example 3: double standard deviation

df3 <- create_error_bar_df(data = data_s, numerical = "sDur", factors = "pauseBin", size = 2)

df3

# example 4: standard error instead of standard deviation

df4 <- create_error_bar_df(data = data_s, numerical = "sDur", factors = "pauseBin", type = "std")

df4

---

Open Exercise {#open_exercise}

The open_exercise function opens the knitted RMarkdown associated with a session of the Statistics for Linguistics Workshop as html file in the system's standard browser.

The function takes a session number as argument. Skipping the zero, e.g. writing 2 instead of 02, works as well.

open_exercise(02)

---

Open Slides {#open_slides}

The open_slides function opens the slides associated with a session of the Statistics for Linguistics Workshop as pdf file in the system's standard browser.

The function takes a session number as argument. Skipping the zero, e.g. writing 1 instead of 01, works as well.

open_slides(01)

---

Compare Variables for Predictor Strength {#predictor_competition}

The predictor_competition function is used to compare the predictive strength of two independent variables. The function creates two identical lmer objects, only differing in fixed effects structure. Then, a log-likelihood test is used to decide which fixed effect structure is better fit to predict the dependent variable.

The function takes a number of arguments as input, i.e. the original data set, the dependent variable for both models, the independent variables to test, a random intercept variable, and (if specified) a random slope variable.

# example 1: two similarly well fit predictors
predictor_competition(data = data_s, dependent = "sDur", 
                      independent1 = "typeOfS", independent2 = "pauseBin", 
                      random.intercept = "speaker")

# example 2: one predictor is better than the other
predictor_competition(data = data_s, dependent = "sDur", 
                      independent1 = "typeOfS", independent2 = "slideNumber", 
                      random.intercept = "speaker")

---

Compute Predictor Strength {#predictor_strength}

This function creates an lmer model for each predictor variable, lacking that predictor variable. Then, conditional and marginal coefficients of determination for each model are calculated. Comparing the value of the conditional coefficient of determination across all models, one can conclude the predictor strength of the respective missing predictor. The function uses MuMIn::r.squaredGLMM to compute coefficients of determination.

This function needs a dependent variable, several independent variables as fixed effects, a random effect structure, and a data set to work.

predictor_strength(dependent = "sDur",
        fixed = c("pauseBin", "list", "folType", "baseDur"),
        random_str = c("(1 | speaker) + (1 | item)"),
        data = data_s)

---

Tukey Contrasts {#tukey}

The tukey function computes Tukey Contrasts for all levels of a categorical predictor.

The function takes simple and multiple linear regression models, as well as linear mixed effects regression models as input. Specify the categorical predictor for which Tukey contrasts should be computed.

simple_lm <- lm(sDur ~ typeOfS, data = data_s)
tukey(model = simple_lm, predictor = typeOfS)

multiple_lm <- lm(sDur ~ typeOfS + pauseBin, data = data_s)
tukey(model = multiple_lm, predictor = pauseBin)

# library(lme4)
mixed_lm <- lmer(sDur ~ typeOfS + pauseBin + folType + (1 | speaker), data = data_s, REML = F)
tukey(model = mixed_lm, predictor = folType)

---

References

Baayen, R. Harald and Shafaei-Bajestan, Elnaz. (2019). languageR: Analyzing Linguistic Data: A Practical Introduction to Statistics. R package version 1.5.0. https://CRAN.R-project.org/package=languageR

Bates, Douglas, Maechler, Martin, Bolker, Ben, and Steve Walker. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-48. doi:10.18637/jss.v067.i01.

Barton, Kamil. (2020). MuMIn: Multi-Model Inference. R package version 1.43.17. https://CRAN.R-project.org/package=MuMIn

Coretta, Stefano, Casillas, Joseph V., and Roettger, Timo. (2021). learnB4SS: Learning materials for the learnB4SS workshop. R package version 1.0.0. https://github.com/learnB4SS/learnB4SS

Hothorn, Torsten, Bretz, Frank, and Westfall, Peter. (2008). Simultaneous Inference in General Parametric Models. Biometrical Journal 50(3), 346-363.

Nakagawa, S., Johnson, P.C.D., and Schielzeth, H. (2017) The coefficient of determination R? and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded. J. R. Soc. Interface 14: 20170213.

R Core Team. (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.

Schmitz, Dominic and Esser, Janina. (2021). SfL: Statistics for Linguistics. R package version 0.2. URL: https://github.com/dosc91/SfL

H. Wickham. (2016) ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York.

---

Please message the author at [e-mail] in case of any questions, errors or ideas.

dosc91/SfL documentation built on Sept. 14, 2024, 6:44 a.m.