2. Definition of survey
Study involving a subset (sample) of
individuals selected from a larger
population
where variables are observed or measured
at a single point in time
aggregated over all individuals in the
sample
to obtain a summary statistic
4. Extension of cross-sectional design
Longitudinal
multiple cross-sectional surveys
can estimate incidence
if results between multiple surveys differ
true difference?
samples very different?
Panel design
multiple cross-sectional surveys using exactly
same sample
Pseudo-cohort
cohort: one age group over time
pseudo-cohort: different age groups studied cross-
sectionally at one point in time
5. Survey
To investigate
if taxi drivers are more likely to use seat belts
than drivers of private automobiles
To determine the community prevalence of
malnutrition
measles
Tuberculosis
HIV
6. Steps in designing a survey
Define survey objectives
Identify study population
Identify and define study variables
Establish plan of analysis
Determine sample size
Choose sampling method
Decide on mode of data collection
Design, field test and revise questionnaire
Train interviewers, conduct field work
Check all interview forms for error
Enter, tabulate and analyse the results
7. Survey objectives
Formulate the objectives of the survey:
What are you interested in finding out?
Who do you want to study?
When do you want to do the survey?
What do you expect to learn and why?
8. Attributes of “SMART” objectives
S pecific
M easurable
A ction oriented
R ealistic
T ime frame
9. Example 1: Objectives
What? What are the knowledge, attitudes
and practices (KAP) to HIV/AIDS
Who? of adults > 18 years of age
Where? in Mwananyamala
When? in year 2006
Why? prior to the start of a new health
education programme
10. Example 2: Objectives
To measure and compare the prevalence of
smoking among MOH and Ministry of
Agriculture employees in Tabora
To measure and compare the proportion of
non-smokers in MOH and MOA exposed to
tobacco smoke in their working environment
and to determine the extent of the exposure
11. Steps in designing a survey
Define survey objectives
Identify study population
Identify and define study variables
Establish plan of analysis
Determine sample size
Choose sampling method
Decide on mode of data collection
Design, field test and revise questionnaire
Train interviewers, conduct field work
Check all interview forms for error
Enter, tabulate and analyse the results
13. Study population
Objectives target population
Study population must be representative of
target population
Time: seasonality, day of the week, time of the
day
Place: urban, rural
Persons: age, sex, other characteristics
A non-representative sample will produce
biased results (validity )
14. Steps in designing a survey
Define survey objectives
Identify study population
Identify and define study variables
Establish plan of analysis
Determine sample size
Choose sampling method
Decide on mode of data collection
Design, field test and revise questionnaire
Train interviewers, conduct field work
Check all interview forms for error
Enter, tabulate and analyse the results
15. Study variables
Identification
as many as necessary but as few as possible
each must “pay its own way”
if of no use in analysis, do not include variable
Requirements
clear definition
method of measuring should be:
repeatable
valid
16. Identify study variables
Objective: To measure and compare the prevalence of
smoking among MOH and MOA employees in Tabora
Variables:
Smokers
in MOH
in MOA
Definition:
A smoker is someone who, at the time of the survey, smokes
cigarettes or any other tobacco products either daily or
occasionally.
daily: smokes any tobacco product ≥ 1 / day
occassionally: smokes, but not every day
17. Steps in designing a survey
Define survey objectives
Identify study population
Identify and define study variables
Establish plan of analysis
Determine sample size
Choose sampling method
Decide on mode of data collection
Design, field test and revise questionnaire
Train interviewers, conduct field work
Check all interview forms for error
Enter, tabulate and analyse the results
18. Establish plan of analysis (POA)
How will the key variable(s) be measured
to meet the objectives?
Proportion, mean, ratio, ...
Level of measurement influences
questionnaire design
type of statistical analysis
What additional variables will need to be
measured?
How will the results be used in the analysis
(dummy tables)?
19. Plan of analysis
To measure and compare the prevalence of smoking among MOH and MOA
staff in Tabora
Variables obtained from all study participants
Number (%) of persons smoking
- daily
- occasionally
Number (%) of persons smoking cigarettes, pipefuls of tobacco, or cigars /
cigarillos
Number (%) of nonsmokers who are
- neversmokers
- exsmokers (and time since quitting)
Variable obtained from smokers
Average number of cigarettes, pipefuls of tobacco, cigars / cigarillos smoked per
day or week
20. Baseline characteristic of study population in UNHCR and UN
--------------------------------------------------------------------------------------------------------
MOH MOA
N = X N = Y
--------------------------------------------------------------------------------------------------------
number (percent)
Age*
30 years or younger
31 - 45 years
46 years or older
Female
Work status
Staff member
Short Term Professional
Short Term Consultant
Others
Dummy table - describe sample
21. Prevalence of smoking among staff of UNHCR and UN
------------------------------------------------------------------------------------------------------
MOH MOA
N = X N = Y
------------------------------------------------------------------------------------------------------
number, percent
Smokers
smoking daily
smoking occasionally
total n = x n = y
Nonsmokers
neversmokers
exsmokers
quit ≤ 1 month ago
quit 2-6 months ago
Dummy table - describe variables
22. Steps in designing a survey
Define survey objectives
Identify study population
Identify and define study variables
Establish plan of analysis
Determine sample size
Choose sampling method
Decide on mode of data collection
Design, field test and revise questionnaire
Train interviewers, conduct field work
Check all interview forms for error
Enter, tabulate and analyse the results
23. Estimating sample size
• Indicate expected frequency of factor of interest
• Decide on desired precision of the estimate
• Decide on acceptable risk that estimate will fall
outside its real population value
• Adjust for estimated design effect
• Adjust for expected response rate
• (Adjust for population size, if sample > 10% of
target population)
24. Methods used in probability samples
Simple random sampling
Systematic sampling
Stratified sampling
Cluster sampling
Multi-stage sampling
25. Steps in designing a survey
Define survey objectives
Identify study population
Identify and define study variables
Establish plan of analysis
Determine sample size
Choose sampling method
Decide on mode of data collection
Design, field test and revise questionnaire
Train interviewers, conduct field work
Check all interview forms for error
Enter, tabulate and analyse the results
26. Method of data collection
Personal interview
Telephone
Self administered
Questionnaire
Mail
Computer direct
E-mail
27. What method to choose?
Nature of questions
What is the research question to be
addressed?
Convenience
Which method is most readily available?
Cost
How much money is available?
28. Personal interviews
“face-to-face”
Advantages
population coverage
can clarify issues, allows complex questions
longer interviews tolerated (2-3 hours)
usually high response rate (~90%)
Disadvantages
expensive
slow
intrusive
requires interviewing skills
interviewer bias
29.
30. Self-administered questionnaires
(mail or hand distributed)
Advantages
cheap
no interviewer bias
long list of response categories possible
sensitive questions
Disadvantages
slow
does not allow clarification
poor response rate (particularly in populations
of lower education and literacy level)
population coverage?
who completes questionnaire??
31.
32. Telephone surveys
“Averages” advantages and disadvantages of
personal and mail surveys
Advantages
relatively fast
sampling through random digit dialling
Disadvantages
need telephone
problem with answering machines, mobile
telephones
no one at home during the day. Limits calling time
to a window of 6-9 p.m.
interview time < 20 minutes
35. Steps in designing a survey
Define survey objectives
Identify study population
Identify and define study variables
Establish plan of analysis
Determine sample size
Choose sampling method
Decide on mode of data collection
Design, field test and revise questionnaire
Train interviewers, conduct field work
Check all interview forms for error
Enter, tabulate and analyse the results
36. Preparing for data analysis
Remember “GIGO” principle
“Garbage in - garbage out”
The computer cannot make data better than
they are
Coding
Data entry
Data cleaning and editing
37. Data analysis
Type of data analysis is dependent on the
of outcome variable
Continuous variables
Categorical variables
38. Points about survey data analysis
Target population in a survey often finite
Most software packages assume:
infinite population sizes
sampling with replacement
both conditions often not upheld in surveys
Software specially developed for surveys
SUDAAN (Survey data analysis)
Stata
Epi Info CSurvey
39. Ethical issues
Informed consent
purpose of study explained?
consent given by individual
to be included?
Confidentiality
an individual should not be identifiable
Service provision
41. Response rate
Dependent on
length and mode of interview
educational level
perceived threat of questions
“warm” versus “cold” interviews
warm = interviewee has prior knowledge about
survey
Non-responders are almost always
different from responders
To reduce non-response bias
follow up non-responders at least once
42. Use of results
Biggest fault in survey design: Results are
not disseminated or used.
Why was the suryey carried out in the first
place?
Editor's Notes
And we could continue this definition with “in order to use the measurements or observations on this sample to draw conclusions regarding the entire population”.
Key phrase: representatives of the entire population
Entire population is not the general population but our population of interest, target population to which the information we obtain is applied.
Sampling, sampe size, questionnaire =&gt; most components to put survey together.
Most importantly, to be absolutely clear about what the objectives of the study are. Often this is the most difficult step in survey design and takes longer than all other steps
often also most frustrating, often only a few lines
Objectives will automatically determine the population to be surveyed.
The general population you want to know about is the target population.
The part of target population that you interview is the sample.
You use the proportion from the sample population to estimate the proportion in the target population.
*Concept of representativity not only to demographic characteristics of pop but to the population WITH regard to the factor studied.
Survey about attitudes towards new unemployment scheme and we conduct this survey between 9 am and 5 pm, we will not get a representative sample of the population.
Those who are at home will be more likely to be unemployed and have a different attitude towards an unemployment policy than the others not at home during workhours. If number of children in HH, the timing may not be so much of an issue. More of concern geographic representativity.
If we want to study factors related to the common cold, it makes a difference it we carry out the study in summer or in winter time.
Attitudes towards f4wheel drive cars, difference f we study Tokyo residents or resident in N Hokkaido
The point with representativity is that our sample has to be representative for the pop. with regard to the factor studied.
Sampling, sampe size, questionnaire =&gt; most components to put survey together.
Failure to establish a PoA is probably the commonest weakness in survey design
will prevent that data is collected that cannot be used later in the analysis, does not contribute to answering the objectives.
Basis for our plan of analysis and our questionnaire design
Failure to establish a PoA is probably the commonest weakness in survey design
will prevent that data is collected that cannot be used later in the analysis, does not contribute to answering the objectives.
Sampling, sampe size, questionnaire =&gt; most components to put survey together.
(1) do not calculate a ss for a study but for a factor of interest. More than 1, may require separate SS calculations. If you want to estimate the prevalence of smoking in a community and determine the morbidity of the identified smokers in the previous year we talk about one study but two different sample size requirement., one dealing with the prevalence of smoking in the community, the other dealing with morbidity among smokers. All the variables that enter into the sample size formula will be different.
3 main methods
can take place in the home, at a shopping mall, on the street, outside a movie theater or polling place, and so on
Population coverage: ability to find the target population. For FP in Suntoko supermarket. Interview people in front
Short list of response categories: at your doctor do you usually have to wait less than 10 minutes, 10-20 minutes, more than 20 minutes
Very popular in US, not so much in Europe, Japan?
Sampling, sampe size, questionnaire =&gt; most components to put survey together.