Healy & Associates
chartered occupational psychologists
Research Design and Statistical Consultancy
Statistics is the science of generalising from a
small sample to a wider population. Anyone who has
ever tried to summarise and draw conclusions from data will appreciate how difficult it
can be. All of our Psychologists are Chartered
Scientists, and good statistical practice is part of
our everyday work, but we also provide a specialist
research consultancy service to clients.
Much of the data analysis we see contains
common errors. The commonest error is failing to apply a
hypothesis driven analysis, and using a grossly
oversimplified (and often completely incorrect)
analysis model. Here are a few of the more common
mistakes we come across :
- Confusing descriptive and inferential data.
- Trying to draw conclusions and make
predictions based on descriptive analysis only.
- Confusing data types - nominal. ordinal,
interval and ratio, and hence using the wrong
- Failing to understand or use a proper
hypothesis testing approach, relying instead on
a 'fishing trip' approach.
- Drawing unfounded conclusions from
- Not understanding the technical definition
of significant, and failing to apply
any statistical technique to test it.
- Making unwarranted extrapolation from weak
- Not being aware of, or understanding Type I or Type II errors.
- Using correlation to infer a causal
- Confusing standardised and raw data types.
- Confusing the nature of, and relationship
between, dependent and independent variables.
- Failing to account for measurement error and
failing to calculate basic descriptive values
such as the Standard Error of the Mean.
- Failing to define analysis as being 'between subjects'
vs. 'within subjects', and hence
being unable to choose
the correct procedure for analysis.
- Not understanding that observed differences
have to be statistically significant,
instead relying on
gut-instinct, or 'eyeball' analysis to conclude
there is a difference, when in fact the data does
not support this.
- Poor quality sampling - non-randomised,
non-stratified, too small, or unrepresentative.
- Failing to adopt good research
practices such as blinding and controlling for
- Not applying Confidence Intervals, or failing to
understand how they affect interpretation of results.
- Carrying out analysis with severely range
There are a lot more but we'll stop the list there!
If you are carrying out analysis of crucial business
information, and will be making decisions based on the
results, we will be happy to help you ensure your
research and analysis model is the correct one, and that
you can have faith in the results.
We can usually provide this service remotely.
We also offer a bespoke Statistics for
Business Training Workshop
designed for people who need a grounding
in basic statistical concepts and procedures.
Please contact us for more details.