What They Say and What They Do: Comparing Physical Activity Across US, UK, and The Netherlands

Arie Kapteyn (USC)(Joint work with Htay-Wah Saw, Andrew Steptoe, Jim Smith,

Arthur van Soest, and many others)

Motivation• Physical activity is a major input in achieving a

healthy life• Increasingly comparative studies across nations

try to assess which policies are most effective in improving the health and well-being of populations

• Largely these studies rely on self-reports of health behavior.

• How comparable are these self-reports?

Our study compares three countries

• Netherlands (NL)• United States (US)• United Kingdom (UK)• Today’s talk is only about NL and US• We use two probability-based Internet panels:• Longitudinal Internet Studies for the Social

Sciences (LISS, NL)• Understanding America Study (UAS, US)

What are accelerometers and how do we use them? Developed by Geneactiv (UK) Measures acceleration, skin temperature, daylight• Design of study in NL: 13 weeks data collection (among members of LISS panel) 70 - 90 panel members per week Panel member wears device for 8 days About 900 LISS respondents• Design of study in England: About 250 respondents of ELSA panel (age 50+) US: Currently about 300 respondents Aiming for 500 respondents in our new Internet panel:

Understanding America Study (UAS)

We do more

• For one weekday and one weekend day we ask about self reported physical activities, as well as global questions about physical activities

Average Total Hours the Devices Were Worn by Dutch Respondents by Day

0

5

10

15

20

25

Hou

rs

Day of Week

Average Total Hours the Devices Were Worn by American Respondents by Day

0

5

10

15

20

25

Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7 Day 8 Day 9 Day 10

Hou

rs

Composition of Dutch and US samplesLISS Population UAS Population

Married 54% 40% 57% 48%Female 51% 50% 55% 51%Low 32% 33% 23% 42%Medium 35% 41% 33% 31%High 33% 25% 43% 27%Working 50% 51% 64% 60%Age(18-39) 21% 34% 29% 36%Age(40-50) 20% 20% 19% 18%Age(51-64) 31% 24% 32% 26%Age(65+) 28% 22% 17% 19%White - 81% 74%Dutch 86% -

How active do respondents say they are?• How in general would you describe the level of your

physical activities?

• Americans may use the extremes of the scale more

LISS UASInactive 8% 10%Mildly active 22% 27%Moderately active 42% 33%Active 24% 20%Very Active 4% 8%Chi-sq (P-value) 16.2 0

If you like pictures….

8

22

42

24

4

10

28

34

21

80

2040

6080

100

Perce

nt

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Chi-sq (P-value) 16.20 0.00

Broken down by marital status

7

25

42

24

2

9

20

42

24

5

13

26

32

19

10

8

29

34

22

6

020

4060

8010

0

Not Married Married Not Married Married

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 8.52 0.07 3.27 0.51

Gender

10

23

43

19

5

6

22

41

28

3

8

24

38

22

8

12

31

30

20

80

2040

6080

100

Male Female Male Female

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 13.77 0.01 4.56 0.34

Education

10

24

45

19

2

9

25

37

24

5

5

17

45

28

5

11

26

33

19

10

11

33

31

21

5

10

25

36

21

90

2040

6080

100

Low Medium High Low Medium High

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 17.15 0.03 3.83 0.87

Age

5

29

38

25

3

10

20

47

17

6

7

23

42

25

3

9

18

43

26

4

11

19

33

23

13

11

31

28

23

7

7

37

32

20

4

16

22

39

18

6

020

4060

8010

0

18-39 40-50 51-64 65+ 18-39 40-50 51-64 65+

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 15.6 0.21 16.75 0.16

Work

8

23

41

26

3

7

22

44

22

5

14

31

33

15

6

8

26

34

24

90

2040

6080

100

Non-working Working Non-working Working

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 4.83 0.31 6.5 0.16

How does this all match up with theAccelerometer Measurements?

To make objective and subjective measures available, we rescaled the objective measures

• We first computed each respondent’s average of the measurements over the observation period– Correcting for how long they wear the device every day

• Next we divide the Dutch data into quintiles– Assign the label “inactive” if the average falls below the 20th

percentile, “mildly active” if between 20th and 40th perc., etc.– By construction, 20% of the Dutch respondents fall in each of

the five activity categories. • Use the Dutch cut-off points also for the US data.• The objective and subjective scales are not comparable,

but we can look at differences across groups.

Comparing the objective measures

20

20

19

20

21

57

15

11

9

8

020

4060

8010

0Pe

rcen

t

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Chi-sq (P-value) 145.83 0

Broken down by marital status

21

18

20

22

19

19

21

18

19

22

59

13

10

7

10

55

16

12

9

7

020

4060

8010

0

Not Married Married Not Married Married

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 2.98 0.56 1.71 0.79

Gender

20

20

20

22

19

20

20

19

19

22

60

12

11

6

11

54

17

12

10

6

020

4060

8010

0

Male Female Male Female

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 1.3 0.86 5.01 0.29

Education

26

21

17

16

19

19

19

20

21

21

15

19

20

24

21

55

11

15

11

8

56

16

10

10

8

58

16

10

7

9

020

4060

8010

0

Low Medium High Low Medium High

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 13.53 0.09 3.5 0.9

Age

9

9

25

26

30

12

22

20

22

23

15

22

20

19

24

39

24

13

16

8

38

24

14

13

10

56

13

16

7

8

62

13

9

10

7

79

6

826

020

4060

8010

0

18-39 40-50 51-64 65+ 18-39 40-50 51-64 65+

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perc

ent

Graphs by liss_uas

Chi-sq (P-value) 104.16 0 29.47 0

Work

29

23

18

16

14

11

17

21

24

27

69

10

8

8

5

50

18

14

9

10

020

4060

8010

0

Non-working Working Non-working Working

LISS UAS

Inactive Mildly active Moderately active ActiveVery active

Perce

nt

Graphs by liss_uas

Chi-sq (P-value) 57.9 0 12.46 0.01

Multivariate Analysis• When conducting multivariate analyses to

explain the pattern of subjective and objective measures of physical activity we find qualitatively similar patterns.

• How much of the observed differences are due to different sample compositions, and how much is due to different behavior of otherwise identical individuals in the two countries?

Oaxaca Decomposition

• We can decompose the difference in a number of ways:

• The first term corrects for the difference in sample composition; the second term reflects how behavior is different for people with the same characteristics

' , ( ) 0, ( , )lY X E A Bβ ε ε= + = ∈

{ } { }( ) ( )( ) ( ) ' ( ) '( ) ( ) ( ) '( )

A B

A B B B A B A B A B

E Y E YE X E X E X E X E Xβ β β β β

− =

− + − + − −

We are comparing regressions in two groups, A and B

Endowment

Coefficients

Interaction

Oaxaca Decomposition

DifferentialNational averages p-values

National averages p-values

Prediction_NL 2.94 0.00 3.01 0.00Prediction_US 2.88 0.00 1.96 0.00Difference 0.06 0.43 1.05 0.00

DecompositionEndowments -0.25 0.07 -0.34 0.04Coefficients 0.20 0.08 1.20 0.00Interaction 0.11 0.52 0.19 0.35

Subjective Objective

comparative

NLUKUSA(1)(2)

subjectiveobjectivesubjectiveobjectivesubjectiveobjectivesubjectiveobjective

Female0.373-0.023-0.1300.040medium0.025-0.759

(0.089)***(0.129)(0.127)(0.184)(0.278)(0.366)**

Age (65+)0.158-0.5280.241-0.144-0.853-0.423high0.029-0.583

(0.110)(0.160)***(0.140)*(0.206)(0.386)**(0.507)(0.293)(0.384)

Married0.2810.189-0.0190.099working0.7710.848

(0.090)***(0.131)(0.127)(0.184)(0.259)***(0.341)**

Medium0.052-0.1200.1540.5050.025-0.759white0.2150.185

(0.109)(0.158)(0.199)(0.293)*(0.278)(0.366)**(0.274)(0.361)

High0.4180.1570.2320.8960.029-0.583age_40_50-0.0650.180

(0.107)***(0.155)(0.237)(0.344)***(0.293)(0.384)(0.328)(0.431)

Working0.1810.6180.2720.8660.7710.848age_51_64-0.5670.115

(0.115)(0.168)***(0.155)*(0.226)***(0.259)***(0.341)**(0.285)**(0.371)

Constant2.3212.6662.6812.237age_65_plus-0.853-0.423

(0.135)***(0.197)***(0.252)***(0.369)***(0.386)**(0.507)

N4514512412469192_cons2.7292.777

R-sq0.0850.1280.0280.1220.2380.144(0.365)***(0.479)***

N9192

R-sq0.2380.144

Standard errors in parentheses

="* p

Concluding Remarks

• Objective and subjective measures tell very different stories

• Subjectively, Americans tell us they are about as active as the Dutch; objectively the difference is more than one point on a 1-5 scale.

• Comparisons within countries based on self-reports are also likely to be misleading– E.g. across age groups; working versus non-working

Slide Number 1MotivationOur study compares three countriesWhat are accelerometers and how do we use them?We do moreAverage Total Hours the Devices Were Worn by Dutch Respondents by DayAverage Total Hours the Devices Were Worn by American Respondents by DayComposition of Dutch and US samplesHow active do respondents say they are?If you like pictures….Broken down by marital statusGenderEducationAgeWorkSlide Number 16To make objective and subjective measures available, we rescaled the objective measuresComparing the objective measuresBroken down by marital statusGenderEducationAgeWorkMultivariate AnalysisOaxaca DecompositionOaxaca DecompositionConcluding Remarks