A Thesis Case Study: Does Social Media Use Affect Student GPA? (Correlation & Regression)

This is a walkthrough of a real master's thesis analysis we supported: a student investigating whether daily social media use predicts academic performance among university students. We use it as a teaching example because it touches almost every step a quantitative dissertation needs — scale reliability, correlation, multiple regression, assumption checks, and APA reporting. If you are doing your own thesis, you can follow this structure end to end.

The Research Question and Hypotheses

• RQ: Does daily social media use predict GPA, controlling for study hours and sleep?
• H1: Higher daily social media use is associated with lower GPA.
• H2: Social media use predicts GPA over and above study hours and sleep.

Writing directional hypotheses up front determines whether you use one- or two-tailed tests later. H1 is directional (we expect a negative relationship).

The Sample and Variables

• N = 312 undergraduate students, collected via an online survey.
• GPA: self-reported, 0.0–4.0 (the outcome / dependent variable)
• SocialMediaUse: average daily hours on social platforms
• StudyHours: average daily hours studying
• SleepHours: average nightly sleep
• PhoneAnxiety: a 6-item Likert scale (1–5), measuring distress when separated from the phone

Step 1: Check the Scale's Reliability First

Before using the PhoneAnxiety scale, we confirmed its items hang together. A scale with poor reliability produces meaningless results.

Analyze > Scale > Reliability Analysis

• Move the 6 items in, statistic = Cronbach's Alpha, and request Scale if item deleted.
• Cronbach's α = 0.84 — above the 0.70 threshold, so the scale is internally consistent. No item improved alpha if deleted, so all six were kept and averaged into a single score.

Step 2: Descriptive Statistics and Screening

Analyze > Descriptive Statistics > Descriptives gave the means and SDs for the table every results chapter needs:

Variable	Mean	SD
GPA	3.12	0.51
Social media (hrs/day)	4.3	2.1
Study hours/day	2.8	1.4
Sleep (hrs/night)	6.6	1.2

We also screened for impossible values (a GPA of 4.7, a 30-hour study day) and removed three data-entry errors.

Step 3: Correlation Analysis

To test H1 and see how variables relate, we ran Pearson correlations:

Analyze > Correlate > Bivariate → tick Pearson and Flag significant correlations.

Pair	r	p
Social media ↔ GPA	−0.38	<.001
Study hours ↔ GPA	0.45	<.001
Sleep ↔ GPA	0.21	<.001
Social media ↔ Sleep	−0.29	<.001

Interpretation: social media use had a moderate negative correlation with GPA (r = −0.38) — H1 is supported at the bivariate level. Students who used more social media tended to have lower GPAs and less sleep. But correlation cannot tell us whether social media matters once we account for study and sleep. That needs regression.

Step 4: Multiple Regression

Analyze > Regression > Linear

• Dependent: GPA
• Independents: SocialMediaUse, StudyHours, SleepHours
• Under Statistics: tick Estimates, Model fit, R squared change, Collinearity diagnostics, Durbin-Watson
• Under Plots: put ZRESID on the Y axis and ZPRED on the X axis, and tick Histogram and Normal probability plot — these check the assumptions.

Checking Assumptions (Do This Before Trusting Any Coefficient)

• Independence of residuals: Durbin-Watson = 1.92 (close to 2 — fine).
• Multicollinearity: all VIF values under 1.4 (well below 10 — predictors aren't redundant).
• Normality of residuals: the normal P-P plot points hugged the diagonal line.
• Homoscedasticity: the ZRESID vs ZPRED scatter showed a random cloud with no funnel shape.

All assumptions held, so the model is trustworthy.

The Model Output

Model Summary:

• R = 0.56, R² = 0.31 — the three predictors together explain 31% of the variance in GPA.
• ANOVA table: F(3, 305) = 45.8, p < .001 — the model is significant overall.

Coefficients table (the heart of the analysis):

Predictor	B	β (Beta)	t	p
(Constant)	2.41	—	9.8	<.001
Social media	−0.052	−0.21	−4.1	<.001
Study hours	0.118	0.32	6.0	<.001
Sleep	0.041	0.10	2.0	.048

Reading it:

• Social media, β = −0.21, p < .001: even after controlling for study hours and sleep, each extra hour on social media predicted a 0.052 drop in GPA. H2 is supported — social media has a unique effect, not just a side-effect of less studying.
• Study hours was the strongest predictor (β = 0.32). Use the standardized beta, not B, to compare which predictor matters most, because the variables are on different scales.

Step 5: Write It Up in APA Format

This is the sentence structure examiners expect:

A multiple regression was conducted to predict GPA from daily social media use, study hours, and sleep. The model was significant, F(3, 305) = 45.8, p < .001, and explained 31% of the variance (R² = .31). Social media use was a significant negative predictor of GPA, β = −0.21, t(305) = −4.1, p < .001, even after controlling for study hours and sleep.

Lessons for Your Own Thesis

• Reliability before analysis. Never run correlations on a multi-item scale without first reporting Cronbach's alpha.
• Correlation is the appetizer, regression is the main course. A bivariate r can vanish once you control for confounds. Regression answers "does it matter on its own?"
• Check assumptions and report that you did. Examiners look for the Durbin-Watson, VIF, and residual plots. A perfect-looking model with unchecked assumptions is a red flag.
• Self-reported data has limits. GPA and screen time were self-reported, so the discussion chapter must acknowledge recall bias. State limitations honestly — it strengthens, not weakens, a thesis.
• Causation caveat. This is correlational. We can say social media predicts lower GPA; we cannot say it causes it. Be precise with that language throughout.

---

Working on a dissertation and stuck on the analysis chapter? Insighter Digital handles the full quantitative workflow in SPSS — reliability, correlation, regression, assumptions, and APA-ready tables — and explains every result so you can defend it. Get help with your thesis.