Quantitative Data Preparation Alasdair Crockett, ESDS Data Services Manager

1
Quantitative DataPreparationAlasdair
Crockett, ESDSData Services Manager
2
What characterises a good quantitative dataset?

• i) Accurate data
• ii) Well labelled data
• iii) Well documented data
• iv) Data that can be stored in user-friendly
  dissemination formats, but can also be archived
  in a future-proof preservation format

3
Accuracy of data validation checks

• Computer aided surveys (CAPI, CATI or CAWI)
• ? These are the most accurate way of gathering
  survey data, but the software (e.g. Blaise) and
  hardware (e.g. a laptop for every interviewer)
  may be beyond project resources 
• ? Computer aided surveys allow one to build in as
  many logical checks - on question routing and
  responses - as is possible at the point of data
  creation.
• Non computer aided surveys
• ? Less control over initial responses, but checks
  can performed
• At the point of data entry/transcription if
  data entry software is used. However, there are
  few cheap data entry packages around.
• The only feasible option may be to enter data
  without checks directly into a spreadsheet style
  interface (e.g. Excel worksheet, SPSS data view),
  and perform validation checks afterwards - via
  command files in statistical packages or Visual
  Basic code in Excel or Access

4
An example of data seemingly untouched by the
human eye

• Originating error in text variables
• Occupation Description of Occupation
• sole trader purveyor of seafood
• Propagated error in derived numeric variables
• Respondent was coded under the standard
  occupational (SIC) code relating to food
  retailers
• 52.2 Retail sale of food, beverages and tobacco
  in specialised stores

5
Labelling of data I

• All variables should be named. Variable names
  should not exceed 8 characters where possible, as
  the most common format for disseminating data is
  SPSS.
• All variables should be labelled. Labels should
  be brief (preferably lt 80 characters), but
  precise and always make explicit the unit of
  measurement for continuous (interval) variables.
  Where possible, all variable labels should
  reference the question number (and if necessary
  questionnaire). For example, the variable
  q11bhexc might have the label q11b hours spent
  taking physical exercise in a typical week. This
  gives the unit of measurement and a reference to
  the question number (q11b), so the user can
  quickly and easily cross-reference to it.

6
Labelling of data I

• For categorical variables, all codes (values)
  should be given a brief label (preferably lt 60
  characters). For example, p1sex (gender of person
  1) might have these value labels 1 male, 2
  female, -8 dont know, -9 not answered. 
• Where possible, all such labelling should be
  created and supplied to the UKDA as part of the
  data file itself. This is the expectation with
  data supplied in one of the three major
  statistical packages - SPSS, STATA or SAS.

7
Documentation

• Core documentation
• ? Questionnaire.
• ? Methodology details of sample design, response
  rate, etc.
• ? Codebook, i.e. a comprehensive list of
  variable names, variable descriptions, code names
  and variable formatting information. This is
  essential If the package being used for data
  management does not allow the sort of variable
  and code labelling to be stored within the data
  file
• ? Technical report describing the research
  project.
• Other useful documentation that is seldom
  supplied
• ? Code used to create derived variables or check
  data (e.g. SPSS, STATA or SAS command files).

8
Good and bad data documentation formats

• For full details for all types of data see
• http//www.data-archive.ac.uk/depositingData/howto
  Deposit.aspformat