18 JASP Workshop - Multiple regression
JASP is a free, open-source statistical software package with a user-friendly, point-and-click interface suitable for research in psychology. It offers a wide range of statistical analyses, from basic descriptive statistics to advanced methods like regression and ANOVA.
JASP is available on most of the university computers, however, we recommend that you also install a personal version for your own laptop or desktop computer so that you can continue learning outside of class time.
JASP can be downloaded here: www.jasp-stats.org/
18.1 Multiple regression - Example analysis
In this video I go through the analysis and answers for Multiple Regression Exercise 1. If you would like to follow along (or have a go at the exercise yourself first) the question sheet and data can be found in the Week 2 module area for NS5108.
18.2 "Where to click" guide - Multiple Regression
Open JASP.
Load the Data: click on the File tab at the top left and select Open. Then navigate to the folder containing your data file and open it.
Identify which of your variable will be your predictors and which will be the dependent variable (the variable you are aiming to predict).
Check linear relationships between predictor variables: Click Regression -> Correlation from the top bar. Move your predictor variables and your dependent variable over to the Variables Box. Click the Scatter plots options. Visually interpret the Correlation plots to determine if all the predictor variables have a linear relationship to the dependent variable.
Check for multicollinearity: Within the same results section from step 3 check the correlation table. Interpret if the correlation coefficients between your predictor variables is large enough to indicate an issue with multicollinearity.
Check normality (of residuals): Click Regression -> Linear Regression from the top bar. Pick your dependent variable and place this in the Dependent variable box. Pick your predictor variables and place them in the Covariates box. In the Plots drop down select Residuals histogram. Visually interpret this histogram for an indication of this assumption check.
Check Homoscedasticity: Also in the Plots drop down, select Residuals vs. predicted. Visually interpret this plot to assess Homoscedasticity.
Extract the Adjusted \(R^2\) value: This should be in the first table of the Linear Regression section.
Determine the significance of the model: This can be obtained though the ANOVA table. F(df,df)=F-value, p=p-value.
Extract values for your model regression equation: This will be the unstandardized values in the coefficients table. See below for how to report your findings.
18.3 JASP lab exercises
There are two additional exercises on Moodle. Each exercise question sheet comes with a dataset and answers sheet. Work through the question and then check your answers with the answer sheet.
18.4 APA Style Guide for Multiple Regression
18.4.1 Hypothesis for Multiple Regression Analysis
Null Hypothesis (H0): Neither self-esteem nor study hours are significant predictors of academic performance.
Alternative Hypothesis (H1): At least one of self-esteem or study hours is a significant predictor of academic performance.
18.4.2 Reporting Multiple Regression in APA Style
A multiple linear regression was conducted to explore the impact of self-esteem and study hours on academic performance. Assumptions of linearity, independence, multicollinearity, and normality were checked and met.
The multiple regression model significantly predicted academic performance, F(2, 97) = 36.8, p < .001, \(R^2\) = .43.
The regression equation was found to be: Academic Performance = 45.2 + 4.1 * Self-Esteem + 5.2 * Study Hours.
Both self-esteem and study hours significantly added to the prediction of academic performance. Specifically, self-esteem was a significant predictor, t(97) = 2.9, p = .005, and contributed to an increase of 4.1 points for every unit increase in self-esteem. Study hours were also a significant predictor, t(97) = 5.1, p < .001, and contributed to an increase of 5.2 points for every additional hour spent studying.
Overall, the model explained approximately 43% of the variance in academic performance (R² = .43).
As such, the null hypothesis can be rejected, indicating that at least one of the predictors significantly impacts academic performance.