poster - Stanford Computer Science
An Analysis of Exercising Behavior in Online Populations David Jurgens, James McCorriston, Derek Ruths [email protected], [email protected], [email protected]
Introduction and Motivation
Data
Exercise is an important component of health. Previous studies of exercise and health have been stymied by the lack of accurate records of exercising activities. In our study, we overcome this limitation by using data from Fitocracy, a gamified workout tracking site that precisely records users' workout histories using over 1,000 unique exercises and where users self-report their age, gender, and height. We analyze nearly half a million users' histories to identify exercising behavior trends and how those behaviors vary by age and gender.
All of a user's activities, profile, social data, and group memberships on Fitocracy were crawled over a six-month period to acquire the complete profiles and workout histories of 441,034 users. Ultimately, 188,265 users recorded at least one workout, with the total dataset comprising 3,130,276 workouts (14.3M activities) over nearly a four year span from February 2011 to January 2015.
How does exercising behavior vary with age and gender?
Methodology: The activities performed by different age groups provide a insightful view into how exercise behavior changes over time. We divided users by gender and into seven age ranges. Within each gender-age cohort, we computed the probability that a user in that cohort records each exercise and then sorted all exercises according to their average probability of being performed. Below, we show the ten exercises that are most likely to be performed by an individual from each cohort. Ages 20-25 Walking Running Elliptical Trainer Crunch Running (treadmill) Push-Up Plank Body Weight Squat Cycling (stationary) Stretching
Ages 25-30 Walking Running Elliptical Trainer Running (treadmill) Push-Up Plank Crunch Body Weight Squat Cycling (stationary) Stretching
Ages 30-35 Walking Running Elliptical Trainer Running (treadmill) Push-Up Plank Crunch Body Weight Squat Walking (treadmill) Cycling
Ages 35-40 Walking Running Elliptical Trainer Running (treadmill) Walking (treadmill) Push-Up Body Weight Squat Plank Crunch Cycling (stationary)
Ages 40-50 Walking Running Elliptical Trainer Walking (treadmill) Running (treadmill) Push-Up Plank Cycling (stationary) Crunch Stretching
Ages 50+ Walking Walking (treadmill) Running Elliptical Trainer Cycling (stationary) Walked up the stairs Cycling Swimming Running (treadmill) Push-Up
Push-Up Barbell Squat Barbell Bench Press Running Barbell Deadlift Pull-Up Dumbbell Bicep Curl Walking Crunch Overhead Press
Push-Up Running Barbell Squat Barbell Bench Press Barbell Deadlift Pull-Up Dumbbell Bicep Curl Walking Crunch Sit-Up
Running Push-Up Barbell Squat Barbell Bench Press Walking Barbell Deadlift Pull-Up Running (treadmill) Cycling Dumbbell Bicep Curl
Running Push-Up Walking Barbell Squat Barbell Bench Press Running (treadmill) Cycling Elliptical Trainer Barbell Deadlift Pull-Up
Running Walking Push-Up Running (treadmill) Cycling Barbell Bench Press Elliptical Trainer Barbell Squat Cycling (stationary) Pull-Up
Running Walking Push-Up Cycling Running (treadmill) Elliptical Trainer Cycling (stationary) Barbell Bench Press Barbell Squat Crunch
Walking Running Cycling Push-Up Elliptical Trainer Cycling (stationary) Running (treadmill) Walking (treadmill) Light Walking (secondary) Walked up the stairs
Men
Women
Ages 0-20 Walking Running Crunch Plank Body Weight Squat Push-Up Elliptical Trainer Running (treadmill) Jumping Jacks Sit-Up
Women's exerce selection focuses on bodyweight movements and cardio and is remains largely similar across age cohorts. Men's exercise selection initially focuses on strength training but then converges with women's exercise choices.
What exercising behaviors are practiced by subpopulations?
30
40
50
60
20
30
Age
40 Age
50
60
Male Female
20
30
40 Age
50
60
Male Female
200 150 100 50 0 20
30
40 Age
50
60
Top Exercises Plank Dumbbell Lunges Dumbbell Side Bend Exercise Ball Crunch Standing Dumbbell Shoulder Press
250
Male Female
200 150 100 50 0
20
30
40 Age
50
Top Exercises Stretching Foam Rolling Shoulder Dislocation Other Bodyweight Band Pull Aparts
Relative Probability Mass
250
Relative Probability Mass
Relative Probability Mass
Our analysis found more behaviors that were practiced by more women than men, suggesting more nuance in female exercising habits. The behaviors here are ordered from mostly-female practiced to mostly-male practiced.
60
300
200 150 100 50 0 20
30
Male Female
20
30
40 Age
50
60
40 Age
50
60
Top Exercises Cycling (stationary) General Insanity Rowing (machine) Stair Machine Walking (treadmill)
400 350 300 250 200 150 100 50 0
Male Female
20
30
40 Age
50
60
100 0 20
30
40 Age
50
60
100 50 0 20
30
40 Age
50
60
1000 800 600 400 200 0 20
20
30
40 Age
50
60
350
Male Female
300 250 200 150 100 50 0 20
30
40 Age
50
60
500 450 400 350 300 250 200 150 100 50 0
Top Exercises Machine Chest Fly Machine Ab Crunch Lat Pulldown Leg Press Machine Shoulder Press Relative Probability Mass
Top Exercises Cycling Swimming Football (US Soccer) Basketball Hiking
Relative Probability Mass
Relative Probability Mass
Male Female
40 Age
50
60
Male Female
250 200 150 100 50 0 20
30
40 Age
Behavior adoption time changes by gender: Women most adopt this behavior in their early 20s, while men adopt it near 30.
The behavior associated with CrossFit-like exercises had an even gender split. 400 350 300 250 200 150 100 50 0
30
Top Exercises Push-Up Body Weight Squat Crunch Plank Sit-Up
300
Male Female
20
30
60
Top Exercises Walking Walked up the stairs Light Walking Yard Work Other Cardio
60
Top Exercises Two-Arm Kettlebell Swing Goblet Squat (kettlebell) One-Arm Kettlebell Swing Body Weight Inverted Row Push-Up
Male Female
20
30
40 Age
50
40 Age
50
60
Male Female
20
30
40 Age
50
60
350
Male Female
20
30
40 Age
50
60
Top Exercises Dumbbell Bicep Curl Standing Dumbbell Shoulder Press Dumbbell Side Lateral Raise Dumbbell Bench Press Dumbbell Flyes
Surprisingly, we found that the proportion of behaviors did not radically change between age cohorts, which suggests people find their preferred activities and stick with them as they grow older.
300
Top Exercises 250 Barbell Squat 200 Barbell Bench Press 150 100 Barbell Deadlift 50 Standing Barbell 0 Shoulder Pres Bent Over Barbell Row
50
Top Exercises 300 250 Leg Press 200 Lat Pulldown 150 Rowing (machine) 100 Seated Cable Row 50 0 Leg Extensions Relative Probability Mass
200
150
Male Female
200 180 160 140 120 100 80 60 40 20 0
Top Exercises Running (treadmill) Walking (treadmill) Boxing Brazilian Jiu-Jitsu Muay Thai
Male Female
20
30
40 Age
50
60
800 700 600 500 400 300 200 100 0
Top Exercises Pull-Up Chin-Up Push-Up Dips - Triceps Ver. Wide-Grip Pull-Up Relative Probability Mass
300
200
Top Exercises Burpee Rowing (machine) Pull-Up Front Barbell Squat Box Jumps
1200
Relative Probability Mass
400
Male Female
Top Exercises 600 500 Standing Calf Raise 400 Walking 300 Abductor Machine200 Adductor Machine100 0 Glute Kickback
Some behaviors buck gendered expectations!
Relative Probability Mass
500
Top Exercises Running Trail Running Walking General P90X Running (Sprints)
250
Relative Probability Mass
Male Female
Relative Probability Mass
Relative Probability Mass
Height influences exercise selection: Men and women practicing the bottom behavior were 1.5in and 0.34in taller than those for the top behavior! 600
700
Age matters: These two behaviors were associated with the oldest individuals on average Male Female
250
Top Exercises Elliptical Trainer General Yoga Pilates Vinyasa Bikram / Hot Yoga
2000 1800 1600 1400 1200 1000 800 600 400 200 0
Relative Probability Mass
Top Exercises Bicycle Crunch Reverse Crunch Plank Russian Twist
500 450 400 350 300 250 200 150 100 50 0
Relative Probability Mass
Male Female
Relative Probability Mass
20
Top Exercises Boot Camp Spinning Arnold's 1% Circuit Zumba Kickboxing
1000 900 800 700 600 500 400 300 200 100 0
Relative Probability Mass
Male Female
Relative Probability Mass
400 350 300 250 200 150 100 50 0
Relative Probability Mass
Relative Probability Mass
Methodology: To identify the underlying behaviors from people's activities, we train a Latent Dirichlet Allocation (LDA) model on users’ exercising data. We model each individual’s history as a reflection of that person engaging in just a few behaviors, where a behavior selects for certain exercises with higher frequency. Below, we show the demographics of the population engaging in the behaviors identified from a 20-behavior LDA-model.
Male Female
20
30
40 Age
50
60
Top Exercises Incline Dumbbell Bench Press Barbell Bench Press Barbell Incline Bench Press Dumbbell Side Lateral Raise Seated Cable Row
Check out our interactive behavior demo at http://networkdynamics.org/resources/exercise
Introduction and Motivation
Data
Exercise is an important component of health. Previous studies of exercise and health have been stymied by the lack of accurate records of exercising activities. In our study, we overcome this limitation by using data from Fitocracy, a gamified workout tracking site that precisely records users' workout histories using over 1,000 unique exercises and where users self-report their age, gender, and height. We analyze nearly half a million users' histories to identify exercising behavior trends and how those behaviors vary by age and gender.
All of a user's activities, profile, social data, and group memberships on Fitocracy were crawled over a six-month period to acquire the complete profiles and workout histories of 441,034 users. Ultimately, 188,265 users recorded at least one workout, with the total dataset comprising 3,130,276 workouts (14.3M activities) over nearly a four year span from February 2011 to January 2015.
How does exercising behavior vary with age and gender?
Methodology: The activities performed by different age groups provide a insightful view into how exercise behavior changes over time. We divided users by gender and into seven age ranges. Within each gender-age cohort, we computed the probability that a user in that cohort records each exercise and then sorted all exercises according to their average probability of being performed. Below, we show the ten exercises that are most likely to be performed by an individual from each cohort. Ages 20-25 Walking Running Elliptical Trainer Crunch Running (treadmill) Push-Up Plank Body Weight Squat Cycling (stationary) Stretching
Ages 25-30 Walking Running Elliptical Trainer Running (treadmill) Push-Up Plank Crunch Body Weight Squat Cycling (stationary) Stretching
Ages 30-35 Walking Running Elliptical Trainer Running (treadmill) Push-Up Plank Crunch Body Weight Squat Walking (treadmill) Cycling
Ages 35-40 Walking Running Elliptical Trainer Running (treadmill) Walking (treadmill) Push-Up Body Weight Squat Plank Crunch Cycling (stationary)
Ages 40-50 Walking Running Elliptical Trainer Walking (treadmill) Running (treadmill) Push-Up Plank Cycling (stationary) Crunch Stretching
Ages 50+ Walking Walking (treadmill) Running Elliptical Trainer Cycling (stationary) Walked up the stairs Cycling Swimming Running (treadmill) Push-Up
Push-Up Barbell Squat Barbell Bench Press Running Barbell Deadlift Pull-Up Dumbbell Bicep Curl Walking Crunch Overhead Press
Push-Up Running Barbell Squat Barbell Bench Press Barbell Deadlift Pull-Up Dumbbell Bicep Curl Walking Crunch Sit-Up
Running Push-Up Barbell Squat Barbell Bench Press Walking Barbell Deadlift Pull-Up Running (treadmill) Cycling Dumbbell Bicep Curl
Running Push-Up Walking Barbell Squat Barbell Bench Press Running (treadmill) Cycling Elliptical Trainer Barbell Deadlift Pull-Up
Running Walking Push-Up Running (treadmill) Cycling Barbell Bench Press Elliptical Trainer Barbell Squat Cycling (stationary) Pull-Up
Running Walking Push-Up Cycling Running (treadmill) Elliptical Trainer Cycling (stationary) Barbell Bench Press Barbell Squat Crunch
Walking Running Cycling Push-Up Elliptical Trainer Cycling (stationary) Running (treadmill) Walking (treadmill) Light Walking (secondary) Walked up the stairs
Men
Women
Ages 0-20 Walking Running Crunch Plank Body Weight Squat Push-Up Elliptical Trainer Running (treadmill) Jumping Jacks Sit-Up
Women's exerce selection focuses on bodyweight movements and cardio and is remains largely similar across age cohorts. Men's exercise selection initially focuses on strength training but then converges with women's exercise choices.
What exercising behaviors are practiced by subpopulations?
30
40
50
60
20
30
Age
40 Age
50
60
Male Female
20
30
40 Age
50
60
Male Female
200 150 100 50 0 20
30
40 Age
50
60
Top Exercises Plank Dumbbell Lunges Dumbbell Side Bend Exercise Ball Crunch Standing Dumbbell Shoulder Press
250
Male Female
200 150 100 50 0
20
30
40 Age
50
Top Exercises Stretching Foam Rolling Shoulder Dislocation Other Bodyweight Band Pull Aparts
Relative Probability Mass
250
Relative Probability Mass
Relative Probability Mass
Our analysis found more behaviors that were practiced by more women than men, suggesting more nuance in female exercising habits. The behaviors here are ordered from mostly-female practiced to mostly-male practiced.
60
300
200 150 100 50 0 20
30
Male Female
20
30
40 Age
50
60
40 Age
50
60
Top Exercises Cycling (stationary) General Insanity Rowing (machine) Stair Machine Walking (treadmill)
400 350 300 250 200 150 100 50 0
Male Female
20
30
40 Age
50
60
100 0 20
30
40 Age
50
60
100 50 0 20
30
40 Age
50
60
1000 800 600 400 200 0 20
20
30
40 Age
50
60
350
Male Female
300 250 200 150 100 50 0 20
30
40 Age
50
60
500 450 400 350 300 250 200 150 100 50 0
Top Exercises Machine Chest Fly Machine Ab Crunch Lat Pulldown Leg Press Machine Shoulder Press Relative Probability Mass
Top Exercises Cycling Swimming Football (US Soccer) Basketball Hiking
Relative Probability Mass
Relative Probability Mass
Male Female
40 Age
50
60
Male Female
250 200 150 100 50 0 20
30
40 Age
Behavior adoption time changes by gender: Women most adopt this behavior in their early 20s, while men adopt it near 30.
The behavior associated with CrossFit-like exercises had an even gender split. 400 350 300 250 200 150 100 50 0
30
Top Exercises Push-Up Body Weight Squat Crunch Plank Sit-Up
300
Male Female
20
30
60
Top Exercises Walking Walked up the stairs Light Walking Yard Work Other Cardio
60
Top Exercises Two-Arm Kettlebell Swing Goblet Squat (kettlebell) One-Arm Kettlebell Swing Body Weight Inverted Row Push-Up
Male Female
20
30
40 Age
50
40 Age
50
60
Male Female
20
30
40 Age
50
60
350
Male Female
20
30
40 Age
50
60
Top Exercises Dumbbell Bicep Curl Standing Dumbbell Shoulder Press Dumbbell Side Lateral Raise Dumbbell Bench Press Dumbbell Flyes
Surprisingly, we found that the proportion of behaviors did not radically change between age cohorts, which suggests people find their preferred activities and stick with them as they grow older.
300
Top Exercises 250 Barbell Squat 200 Barbell Bench Press 150 100 Barbell Deadlift 50 Standing Barbell 0 Shoulder Pres Bent Over Barbell Row
50
Top Exercises 300 250 Leg Press 200 Lat Pulldown 150 Rowing (machine) 100 Seated Cable Row 50 0 Leg Extensions Relative Probability Mass
200
150
Male Female
200 180 160 140 120 100 80 60 40 20 0
Top Exercises Running (treadmill) Walking (treadmill) Boxing Brazilian Jiu-Jitsu Muay Thai
Male Female
20
30
40 Age
50
60
800 700 600 500 400 300 200 100 0
Top Exercises Pull-Up Chin-Up Push-Up Dips - Triceps Ver. Wide-Grip Pull-Up Relative Probability Mass
300
200
Top Exercises Burpee Rowing (machine) Pull-Up Front Barbell Squat Box Jumps
1200
Relative Probability Mass
400
Male Female
Top Exercises 600 500 Standing Calf Raise 400 Walking 300 Abductor Machine200 Adductor Machine100 0 Glute Kickback
Some behaviors buck gendered expectations!
Relative Probability Mass
500
Top Exercises Running Trail Running Walking General P90X Running (Sprints)
250
Relative Probability Mass
Male Female
Relative Probability Mass
Relative Probability Mass
Height influences exercise selection: Men and women practicing the bottom behavior were 1.5in and 0.34in taller than those for the top behavior! 600
700
Age matters: These two behaviors were associated with the oldest individuals on average Male Female
250
Top Exercises Elliptical Trainer General Yoga Pilates Vinyasa Bikram / Hot Yoga
2000 1800 1600 1400 1200 1000 800 600 400 200 0
Relative Probability Mass
Top Exercises Bicycle Crunch Reverse Crunch Plank Russian Twist
500 450 400 350 300 250 200 150 100 50 0
Relative Probability Mass
Male Female
Relative Probability Mass
20
Top Exercises Boot Camp Spinning Arnold's 1% Circuit Zumba Kickboxing
1000 900 800 700 600 500 400 300 200 100 0
Relative Probability Mass
Male Female
Relative Probability Mass
400 350 300 250 200 150 100 50 0
Relative Probability Mass
Relative Probability Mass
Methodology: To identify the underlying behaviors from people's activities, we train a Latent Dirichlet Allocation (LDA) model on users’ exercising data. We model each individual’s history as a reflection of that person engaging in just a few behaviors, where a behavior selects for certain exercises with higher frequency. Below, we show the demographics of the population engaging in the behaviors identified from a 20-behavior LDA-model.
Male Female
20
30
40 Age
50
60
Top Exercises Incline Dumbbell Bench Press Barbell Bench Press Barbell Incline Bench Press Dumbbell Side Lateral Raise Seated Cable Row
Check out our interactive behavior demo at http://networkdynamics.org/resources/exercise
