Data Editing and Data Quality: Value for Money or Art for Arts Sake The Polish Experience with the E

1
Data Editing and Data QualityValue for Money or
Art for Arts Sake? The Polish Experience with
the ESS

• Zbyszek Sawinski
• Polish Academy of Sciences

PRAGUE June 2528. 2007
2
Outline

• The overview of data processing
• Three types of data comparisons
• Which data are be corrected
• How strongly data editing modifies the data
• Data transposition fieldwork histories
• Conclusions value for money or art for arts sake

3
The overview of data processing
Contact Forms
Sampling data
Questionnaires
DATA ENTRY EDITING
Contact Formsdata file
Sampling file
Questionnairedata file
4
Three types of data comparisons

• Data collected twice
• Interrelated questions
• Data juxtaposed with external sources

5
Data collected twice

• Fieldwork data
• Interviews vs Contact Forms
• respondents ID
• interviewers ID
• date of interview
• start time

6
Interrelated questions
42
7
Data juxtaposed with external sources

• Key demographics
• Gender
• Age
• Size of domicile

8
Which data are to be corrected?

• Contradictory interviewers recordings are to be
  corrected in all cases.
• eg different dates of interview recorded in the
  Questionnaire and in the Contact Form
• eg interview end time comes earlier than
  interview start time
• If there is no basis for correcting both
  variables, then one or both should be coded as
  missing.

9
Which data are to be corrected?

• Contradictory answers come from the respondent
• (1) First check whether or not the error may
  have been committed by the interviewer and if it
  may be corrected
• eg. in the household grid an interviewer recorded
  the same gender for the respondent and for the
  respondents partner.
• (2) When an error cannot be corrected code one
  or both answers as missing
• eg. in the household grid an interviewer recorded
  the current year (2006) as the year of a
  partners birth.

10
To correct or not to correct?

• Respondent refused to answer
• impute data, when it is known
• An answer is inconsistent with external data
• eg year of birth (household grid 1976, sampling
  file 1967)
• consistent with other answers ? do not change
• inconsistent ? change
• We can never exclude a situation that the
  interview was conducted with a person other than
  the selected one, which may have happened either
  by mistake or intentionally.

11
Summary of data corrections
Note Total number of interviews is 1721.
12
Top 10 items most often corrected
13
Transposition of data into fieldwork histories
Interviewer 1
day 1 time a visit/no intrv. time a visit/no
intrv. time an interview day 2 time an
interview time an interview time a visit/no
intrv. time an interview . . . . .
Contact Forms
Interviewer 2
. . . . . day 2 time an interview time a
visit/no intrv. time an interview day 3 time
a visit/no intrv. time a visit/no intrv.
time a visit/no intrv. time an interview .
. . . .
14
Fieldwork history an interviewer following
procedures
15
Fieldwork history an interviewer NOT following
procedures
16
Conclusions

• Opportunities offered by consistency checks are
  significantly limited.
• Most interviewers errors and other types of
  errors cannot be identified during data
  processing. It is too late!
• Data improvement opportunities are limited,
  however...
• 1. One can significantly reduce or even totally
  eliminate missing data and errors pertaining to
  key demographics.
• 2. One can identify at least some questions which
  may be understood ambiguously or which pose other
  types of difficulties.
• Non-standard analyses of fieldwork data can
  providefresh and useful insights into
  interviewers work.

17
DATA EDITING
?
Value for money
Art for Arts sake