Multiple Regression Analysis of Starbucks Prepaid Card Spending
This paper presents a series of multiple regression analyses using customer and sales data from Starbucks. Three dependent variables are examined in turn: the dollar amount loaded onto prepaid cards, the number of days per month customers visit Starbucks, and total sales revenue. Predictors include customer age, gender, income, daily coffee consumption, and operational variables such as number of stores and average weekly earnings. The paper interprets SPSS regression output — including R², F-statistics, standardized beta coefficients, VIF, tolerance, and Durbin-Watson statistics — to evaluate model fit, statistical significance, and multicollinearity. Findings reveal that daily coffee consumption is the strongest significant predictor of prepaid card spending, while the revenue model suffers from severe multicollinearity requiring further refinement.
- Introduction and Overview of Regression Models: Overview of the four regression analyses conducted
- Predicting Starbucks Prepaid Card Amount: Coffee consumption significantly predicts prepaid card spending
- Predicting Days per Month at Starbucks: Cups per day predicts visit frequency; model moderately fits
- Predicting Sales Revenue by Number of Drinks Sold: Revenue model fits perfectly but shows severe multicollinearity
- Contribution of Gender to the Prepaid Card Model: Gender added as predictor; remains non-significant in model
✍️ How to write this paper — guide, tools & examples ▾
What makes this paper effective
- Methodically interprets each SPSS regression output in plain language, translating statistical values (R², F, beta, VIF, Durbin-Watson) into substantive meaning for a non-specialist reader.
- Honestly acknowledges the limitations of each model — weak R², multicollinearity problems, and marginal significance — rather than overstating findings.
- Connects statistical results to real-world business context (e.g., store cannibalization, product diversification), grounding the analysis in practical relevance.
Key academic technique demonstrated
The paper demonstrates systematic diagnostic checking in multiple regression: it evaluates not just the overall model fit (R², F-statistic) but also inspects individual predictor significance, standardized beta coefficients, VIF and tolerance values for multicollinearity, and the Durbin-Watson statistic for autocorrelation. This layered diagnostic approach is essential for credible quantitative analysis and shows the writer understands that a significant F-statistic alone does not validate a regression model.
Structure breakdown
The paper is organized around four separate regression analyses, each introduced with a brief methodological statement, followed by interpretation of the model summary, ANOVA table, coefficients, and collinearity diagnostics. A concluding section adds a fifth-variable model (gender) to the prepaid card analysis to test incremental predictive value. Raw data and full SPSS output tables appear as supporting appendices throughout.
Introduction and Overview of Regression Models
This paper presents a series of multiple regression analyses applied to Starbucks customer and sales data. Four regression models are examined in sequence. The first model predicts the dollar amount customers load onto Starbucks prepaid cards. The second predicts how many days per month customers visit Starbucks. The third explores the relationship between operational variables and total sales revenue. The fourth extends the first model by adding gender as an additional predictor. Each model is evaluated using standard regression diagnostics, including R², the F-statistic, standardized beta coefficients, variance inflation factors (VIF), tolerance values, and the Durbin-Watson statistic.
Predicting Starbucks Prepaid Card Amount
Multiple regression was used to explore how well the dollar amount loaded onto a Starbucks prepaid card can be predicted from other customer variables, and which variables show the most promise for generating a reliable prediction. The results indicated that the four predictors — age, days per month at Starbucks, cups of coffee per day, and income — explained only 27% of the variance (R² = .27, F = 1.881, p > .05). Because the overall model was not statistically significant, these results should be interpreted with caution.
The standardized beta coefficients for the independent variables are as follows: Age, β = .313; Days per Month, β = .362; Cups of Coffee per Day, β = −.520; Income ($1,000), β = .201. Of these, the number of cups of coffee consumed per day was the only variable that significantly predicted the amount of money loaded onto the prepaid card (β = −.520, p < .05). It is worth noting that this significance level just barely reached the threshold at p = .049.
The negative sign of the cups-of-coffee-per-day coefficient is consistent with the theoretical expectation that customers who anticipate consuming more coffee each day will load larger amounts onto their prepaid cards — that is, higher daily consumption is associated with higher card balances, with the direction of the unstandardized coefficient (B = −2.590) reflecting the inverse coding structure of the variable rather than a counterintuitive relationship. The variance inflation factors (VIF) for all variables in this model are below 2.0, and tolerance values are all above 0.5, indicating that multicollinearity is not a concern.
Predicting Days per Month at Starbucks
A second regression model was estimated to predict how many days per month customers visit Starbucks. Prepaid card amounts were excluded from the predictor set in this model. The results indicated that the four predictors — age, gender, cups of coffee per day, and income — explained 47.1% of the variance (R² = .471, F = 4.457, p ≤ .01). This model represents a meaningfully better fit than the prepaid card model, though ideally R² would be in the mid-to-high range to be considered robust.
The standardized beta coefficients are as follows: Age, β = −.138; Gender, β = −.248; Cups of Coffee per Day, β = .516; Income ($1,000), β = .268. The number of cups of coffee consumed per day was a statistically significant predictor of monthly visit frequency (β = .516, p < .01), meaning customers who drink more coffee per day tend to visit Starbucks more frequently each month.
The Durbin-Watson statistic for this model is 1.384, which is below 2, suggesting that autocorrelation is not a substantive problem in this instance. Overall, the model is not a robust fit, but the results do suggest it may be worth further exploring the contributions of gender and age. Additional model specifications could help isolate these effects.
Always verify citation format against your institution’s current style guide requirements.