You are on page 1of 3

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/263098913

Unit of observation versus unit of analysis

Article  in  BMJ Clinical Research · June 2014


DOI: 10.1136/bmj.g3840 · Source: PubMed

CITATIONS READS

20 19,141

1 author:

Philip M. Sedgwick
St George's, University of London
431 PUBLICATIONS   8,598 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

BMJ Statistics Endgames View project

Biologics for Rheumatoid arthritis pain View project

All content following this page was uploaded by Philip M. Sedgwick on 15 January 2018.

The user has requested enhancement of the downloaded file.


BMJ 2014;348:g3840 doi: 10.1136/bmj.g3840 (Published 13 June 2014) Page 1 of 2

Endgames

ENDGAMES

STATISTICAL QUESTION

Unit of observation versus unit of analysis


Philip Sedgwick reader in medical statistics and medical education
Institute of Medical and Biomedical Education, St George’s, University of London, London, UK

Researchers investigated the effects of a school based recruited through cluster sampling.3 Six primary schools in
educational programme aimed at reducing the consumption of southwest England were recruited and 29 classes were involved
carbonated drinks to prevent weight gain in children aged 7-11 in the trial. The class was the cluster, and all children in each
years. A cluster randomised controlled trial study design was selected cluster were invited to participate. The clusters rather
used. The programme, which was delivered over one school than the children were randomised to treatment using cluster
year, focused on promoting a healthy diet. The control group allocation.4 The class was therefore the unit of randomisation.
received no intervention. Six primary schools in southwest The unit of randomisation is defined statistically as the “who”
England were recruited and 29 classes were involved in the trial. or “what” that is randomised to treatment in a trial.5 All the
Classes were randomised to treatment, with 15 classes allocated children in a cluster received the treatment—a school based
to the educational programme (325 children) and 14 to the educational programme or control (no intervention)—that their
control treatment (319 children).1 cluster had been allocated to. The cluster was therefore the unit
The outcome measures included consumption of carbonated of intervention, defined statistically as the “who” or “what” for
drinks. Each child recorded the number of glasses (average size which the intervention was delivered.
250 mL) that he or she drank over a three day period at baseline In the example above, data were recorded for each child. The
and at the end of the trial, and the change from baseline was consumption of carbonated drinks over a three day period at
obtained. For each class (the cluster), the average change in baseline and at the end of the trial was measured. The change
consumption of carbonated drinks across all children was in consumption of carbonated drinks from baseline was obtained
derived. The treatment groups were compared with regard to for each child. Therefore, the unit of observation was the child
the average change within clusters. Over one school year, the (a is false). For each cluster, the mean change in consumption
consumption of carbonated drinks decreased by a mean of 0.6 of carbonated drinks across the children in the class was derived.
glasses per cluster in the intervention group but increased by The treatment groups were compared with regard to the average
0.2 glasses per cluster in the control group (mean difference change across clusters (15 for the intervention and 14 for the
0.7, 95% confidence interval 0.1 to 1.3). control). Therefore, the unit of analysis was the cluster (b is
Which of the following statements, if any, are true? true).
a) The unit of observation was the class In the example above, the unit of analysis was the cluster.
Because each cluster provided only one measurement, the data
b) The unit of analysis was the class
were considered independent, so standard statistical tests could
c) It can be assumed that the measurements for children be used to compare treatment groups. Alternatively, the child
within a class (cluster) were independent of each other could have been the unit of observation. The treatment groups
Answers would then have been compared regarding the change in
consumption of carbonated drinks averaged across the trial
Statement b is true, whereas statements a and c are false. participants. However, if the child had been the unit of analysis,
The unit of observation and unit of analysis are often confused. the probability of spurious significant findings and misleading
The unit of observation, sometimes referred to as the unit of conclusions would have increased. This is because a cluster
measurement, is defined statistically as the “who” or “what” randomised controlled trial design was used—children within
for which data are measured or collected. The unit of analysis a class would be more likely to respond in a similar manner to
is defined statistically as the “who” or “what” for which treatment and could not be assumed to be acting independently
information is analysed and conclusions are made. (c is false). Children within the same cluster would be more
likely to experience similar outcomes than would those in other
The above trial used a cluster randomised controlled trial study
clusters, irrespective of treatment allocation. This lack of
design, described in a previous question.2 Participants were

p.sedgwick@sgul.ac.uk

For personal use only: See rights and reprints http://www.bmj.com/permissions Subscribe: http://www.bmj.com/subscribe
BMJ 2014;348:g3840 doi: 10.1136/bmj.g3840 (Published 13 June 2014) Page 2 of 2

ENDGAMES

independence of measurements within clusters is usually prone to the ecological fallacy. The ecological fallacy is a term
assessed with the intraclass correlation coefficient (ICC), used when collected data are analysed at a group level and the
described in a previous question.6 If the child had been the unit results are assumed to apply to associations at the individual
of analysis then comparison of treatment groups would have level.9
needed to account for the lack of independence between children
within a class (the cluster). Competing interests: None declared.
Consideration of the unit of observation and unit of analysis is
important in other study designs, such as ecological studies, and 1 James J, Thomas P, Cavan D, Kerr D. Preventing childhood obesity by reducing
consumption of carbonated drinks: cluster randomised controlled trial. BMJ 2004;328:1237.
not just clinical trials. Ecological studies were described in a 2 Sedgwick P. Cluster randomised controlled trials. BMJ 2012;345:e4654.
previous question,7 in which the example used investigated the 3
4
Sedgwick P. Cluster sampling. BMJ 2014a;348:g1215.
Sedgwick P. Treatment allocation in trials: cluster randomisation. BMJ 2014b;348:g2820.
association between child wellbeing and economic status in rich 5 Sedgwick P. Clinical trials: units of randomisation. BMJ 2014;348:g3297.
developed societies.8 Twenty three of the richest 50 countries 6 Sedgwick P. Intraclass correlation coefficient. BMJ 2013;346:f1816.
7 Sedgwick P. Ecological studies: advantages and disadvantages. BMJ 2014;348:g2979.
in the world were included. Data were collected for the child 8 Pickett KE, Wilkinson RG. Child wellbeing and income inequality in rich societies: ecological
and aggregated across the country. Therefore, the unit of cross sectional study. BMJ 2007;335:1080.
9 Sedgwick P. The ecological fallacy. BMJ 2011;343:d4670.
observation was the child, whereas the unit of analysis was the
country. Because the unit of observation and unit of analysis
Cite this as: BMJ 2014;348:g3840
are different in ecological studies, results from such studies are
© BMJ Publishing Group Ltd 2014

For personal use only: See rights and reprints http://www.bmj.com/permissions Subscribe: http://www.bmj.com/subscribe
View publication stats

You might also like