Transcript Panel Data Collection through VAAs and their use

What is panel data?
• ”I have data from a web panel” ≠ panel data
• When the same people respond twice (t>1)
• AND at least one variable are measured twice
www.mod.gu.se
www.mod.gu.se
www.mod.gu.se
www.mod.gu.se
www.mod.gu.se
www.mod.gu.se
- traditionally, the main use of panel data has been to improve
causal inference in observational studies
www.mod.gu.se
•But the problems of causal inference also go deeper
•Maybe some Xs are not exogeneous to your Y?
EU
EU
EU
EU
• With T=2 it doesnt matter
• With T>2 long format is often prefered
• Wide format is easier to understand, and
some descriptive statistics is easier to get
• For causal analysis through panel regression
models, long format is necessary
Collection of panel data through VAAs
• The VAAs main function here is to collect e-mailadresses
• So primarily a way to recruit a web panel
• Or simply to start a specific panel data collection
The swedish panel data example
1)
2)
3)
4)
5)
6)
7)
8)
9)
sample recruited in 2009 and 2010
Wave 1 in early 2014
Wave 2 spring 2014
Wave 3 just before the EUP election in May 2014
Wave 4 just after the EUP election in May 2014
Wave 5 before the national election in September 2014
Wave 6 right after the national election in 2014
Wave 7 .... probably in late November 2014
(and all these people can be linked to the VAA user data)
The swedish panel data example
1) New respondents recruited in the two elections in 2014
2) In total approximately 30,000
% of all users leaving their e-mailadress
4.0%
3.5%
3.0%
2.5%
2.0%
1.5%
1.0%
0.5%
0.0%
9000
8000
160000
Extra survey
140000
Surveys completed
6000
nr recruited
120000
nr users
100000
Kompassanvändare
7000
5000
80000
4000
3000
2000
1000
0
60000
40000
20000
0
Response rates in different panel waves
80%
70%
60%
50%
europrofiler
40%
valkompass (vaa)
30%
Aftonbladet 2014
20%
10%
0%
kp1
kp2
kp3
kp4
kp5
kp6
Response rates in a standard panel data project
• An election panel started two years before the
election (2012)
• Total of five waves
• Half probability based recruitment to the Citizen
Panel at UofG
• Half opt-recruits (partly also from VAAs in 2010,
but not only)
• Total approximately 5-6000 respondents
90%
80%
70%
Participation rates
78%
75%
71%
67%
68%
63%
60%
63%
56%
59%
53%
50%
49%
40%
43%
30%
20%
10%
0%
wave 1
wave 2
wave 3
Opt-in
wave 4
Probability
wave 5
Answered
all
60
Number of answered waves
56
50
43
40
30
18
20
10
13
10
9
12
16 15
9
0
1
2
3
Opt-in
Probability
4
5
Political interest
60
50
40
30
20
10
0
wave 1
wave 2
wave 3
wave 4
Very interested
Rather interested
Not very interested
Not interested at all
wave 5
Increased political interest from not interested to interested
4
3.4
3.5
2.9
3
2.5
2.8
2.4
2.2
2
1.5
1.5
1.1
1.1
1
0.5
0
wave 2
wave 3
Opt-in
wave 4
Probability
wave 5
100
Increased political interest from not interested to
interested
94.6
89.3
90
80
70
60
50
40
30
20
10.7
10
5.4
0
0
1
Opt-in
Probability
Swedish economy today
70
60
50
40
30
20
10
0
wave 1
wave 2
wave 3
Good
Bad
wave 4
wave 5
Change in coming year: The Swedish economy
40
35
30
25
20
15
10
5
0
wave 1
wave 2
wave 3
Will improve
wave 4
Get worse
wave 5
4.5
4
Turned positive regarding economic situation in
Sweden
4
3.5
3
3.3
2.9
2.5
2
2
1.5
2
1.8
1.4
1.3
wave 4
wave 5
1
0.5
0
wave 2
wave 3
Opt-in
Probability
3
2.5
Turned negative regarding economic situation in
Sweden
2.3
2.4
2
1.7
1.6
1.5
1.5
1.3
1.2
1.1
1
0.5
0
wave 2
wave 3
Opt-in
wave 4
Probability
wave 5
www.mod.gu.se