Dataset | Project Reports Index ]

The Exercise Group

Sally Barney
Cheri Boyd
Jackie Hall
Jane Matthews

Q: Does caffeine have an effect on recovery from exercise?


  1. Random assignment to caffeine/no caffeine group (avoid cross-over effect).

  2. Record baseline measurements for heart rate, systolic and diastolic blood pressure

  3. Survey (caffeine? how much? when? previous exercise today? when? eat breakfast? usual exercise amount? usual caffeine intake?)

  4. Exercise (walk/run flights of stairs until heartrate is approx. 150 beats per minute)

  5. Record same three measurements at 0, 1, 5, and 10 minute marks after end of exercise

  6. Switch groups, repeat steps 2-5 next day, same time.

Our Analysis


  1. Boxplots and scatterplots.

    We used the differences between each timed measurement and the baseline reading for that measurement.

  2. Connected line plots of mean readings in each category (immediate with caffeine, immediate without caffeine, one minute with caffeine, one minute without caffeine, etc.). These graphs were easier to read:

Regression analysis attempts:

The graphs of the means against time, particularly for the diastolic rates, motivated our search for a regression model. The data appeared to be exponential or quadratic, so we experimented with various transformations: log, log(log), reciprocal, or square root in search of a linear model. When we couldnąt find a linear model, we tried multiple regression in search of a quadratic model: time & time2, time & time2 & caffeine/no caffeine, or time & time2 & caffeine/no caffeine(1 + time + time2).

Comparison of Variance:

The variability seemed greater in the no caffeine data than in the caffeine data in the diastolic study. We first observed this phenomenon in the boxplots, then confirmed it by looking at the descriptive statistics.


  • Increase sample size.

  • Control caffeine intake and timing of exercise in relation to intake.

  • Exercise on a bike or treadmill (muscles quickly gave up on stairs).

  • Continuously monitor heart rate and blood pressure.

  • Find a regression model.


    The data records the differences from the individuals' resting rates at each of times 0,1,5,10 minutes.

    Return to Top

    Link to Accompanying Dataset

    Return to STATS Project Reports Index