Also check whether the variable measured is appropriate for the construct of
interest. For example, the construct ‘social class’ may have been measured by
asking respondents what newspapers they read regularly. Certain newspapers are
given scores which are equated with higher or lower social classes. You should
ask yourself whether this measure of social class is likely to have any construct
validity.
This initial scrutiny of the raw data provides a second opportunity, the first being
at the design stage, to consider whether all the data collected are required for
subsequent statistical analysis. It is remarkable how often researchers collect
information which is not central to the purpose of an investigation. It is preferable
to have a smaller amount of data of high quality than a large amount of ‘dirty’
data, that is data which is incomplete or illegible.
- Once the criteria of utility and appropriateness have been established it is advised to
consider exactly how data were recorded. Ask yourself, were questions ticked or
circled by respondents? Were numeric values entered by the researcher? Consistency
is important. For example, either integers (whole numbers) should be used throughout
(don’t change from case-to-case) or values should be recorded to the same number of
decimal places. Make sure you can distinguish between missing values—no value
recorded, out of range values—a value recorded but known to be impossible, and for
questionnaire data, ‘don’t know’ and ‘not applicable’ responses.
Beware of problems when data from different sources are combined into one data
set. The same variable may have been measured in different ways, for example,
by asking slightly different questions or recording to a different number of
decimal places.
- Consideration of what roles variables have in the overall study design is important. For
example, whether a nominal variable was used as a stratifying factor in a sample
design or whether a continuous variable will be turned into a categorical variable and
used for stratification. A stratification variable or stratifying factor is a variable that
is used to separate the target population into a number of groups or strata where
members of each strata have a common characteristic, such as stratification of
postgraduate students by fee-paying status, stratum i) UK fee-paying status; and
stratum ii) Overseas fee-paying status.
Similarly, a variable may be used as a controlling factor in an experimental
design, as a covariate, or as a blocking variable in a factorial design. The variable
acting as covariate would need to be a continuous measure and the blocking
variable a categorical variable. In some designs it is important to distinguish
between response (outcome) variables and explanatory (independent) variables.
In a regression design, ‘A-level points score’ may be an explanatory variable and
‘degree performance’ the response variable (sometimes called the criterion
variable in regression analysis). More complex experimental designs, such as
repeated measures and nested designs, may require the data to be entered in a
particular format. You should consult appropriate manuals, such as SAS or SPSS
Procedure Guides, for the statistical analysis procedure that you are using.
Statistical analysis for education and psychology researchers 34
