Transcript No Slide Title

Non-Response Bias in Income Data
Mari Toomse
University of Essex
Research topic
• Non-response bias in annual income
measures:
–
–
–
–
Sources and mechanism of bias
Predicting bias from survey data
Relationship with measurement error
Cumulative bias in a panel study
Data
• Estonian EU-SILC 2007 (Statistics on
Income and Living Conditions)
• Panel study on income
• Comes with call record data
• Linked on individual level to Tax and
Customs Board register
– Respondents AND
– Non-respondent
Research design I
• Non-response bias – variable specific
• Response process as a sequence of events
–
–
–
–
–
Case issued
Address known
Contact achieved
Co-operation achieved
Item response
Research design II
• Mechanisms causing these events are
different
• Assess the net effect of each processes
• Annual salary data from the regiser
• Linked to the outcome of the whole of the
selected sample
– New cases
– Panel cases
Mean salary
90,000
88,000
86,000
84,000
EEK
82,000
80,000
78,000
76,000
74,000
72,000
70,000
Total sample,
frame error
excluded
Issued
First year
Address known
Second year
Contact made
Third and fourth year
HH interview
completed
Responded to
the salary
question
Probability of contact, 1st year
Intercept
Salary
Type of settlement
Rural
Area
Northern
Western
Central
Northeastern
Gender
Male
Age
Model 1
2.117**
0.000**
Model 2
2.109**
0.000
Model 3
1.407**
0.000
0.090
0.092
-0.726**
0.640
1.724**
-0.021
-0.744**
0.652
1.735
-0.027
-0.111
0.019**
Probability of item response, 1st
year
Intercept
Salary
Type of settlement
Rural
Area
Northern
Western
Central
Northeastern
Gender
Male
Age
Model 1
3.449**
0.000**
Model 2
3.458**
0.000**
Model 3
3.251**
0.000**
0.383
0.833
0.346
-0.453
1.157
-0.565
0.307
-0.482
1.141
-0.556
-0.433
0.003
Quintile distribution, 1st year
25.0
20.0
%
15.0
10.0
5.0
0.0
Total sample,
frame error
excluded
Issued
Address known
Contact made
Lowest quintile
Highest quintile
HH interview
completed
Responded to
the salary
question
Conclusions
• There is a substantial negative net bias in
annual salary estimates
• Bias accumulates over the course of a panel
• Main sources of bias are non-contact and
item non-response
• Bias due to non-contact is largerly
explained away by basic grographical
variables
Thank you for your attention!