Kevin Beard Muons Inc. Alex Bogacz, Vasiliy Morozov, Yves Roblin Jefferson Lab
How to Design a Sample and Improve Response Rates Alex StannardScottish Government Kevin...
-
Upload
adam-francis -
Category
Documents
-
view
217 -
download
2
Transcript of How to Design a Sample and Improve Response Rates Alex StannardScottish Government Kevin...
How to Design a Sample and Improve Response Rates
Alex Stannard Scottish GovernmentKevin Ralston University of Stirling
Plan
• Types of sample
• Sampling frames
• Improving response rates
• Group exercise
Census or sample?
Edinburgh City Council wants to know peoples thoughts on its leisure facilities.
A decision is made to send a questionnaire to every household in Edinburgh City. A 5% response rate is expected.
This is a census of all households in Edinburgh City
What is Sampling?
• Identify population
• Select members of population to sample
• Study selected members (the sample)
• Draw inferences about population from sample
Types of samples
• Non-Probability SamplesNot all members of the population have a
chance to be included in the sample. Selection method not random.
• Probability Samples Every member of the population has a known,
nonzero chance of being included in the sample. The selection method is random.
Probability Sampling
Example – 1000 households randomly selected across Argyll and Bute.
• More expensive and slower• Non response a problem – but resources can be
targeted and extent of non response bias can be estimated.
• Enables precision of final statistics to be assessed.• Sample selection method is objective, specified and
replicable.
Simple random sampling
• Sampling method completely random based on random numbers.
• Is easy to understand, but can be expensive
Example – every household in Aberdeen City assigned a number. 1,000 random numbers chosen between 1 and 204,683. These numbers identify which households are in sample.
• Tables of random numbers• www.random.org• Excel function ‘=rand()’
Systematic Sampling
• Uses a ‘random’ start on the sampling frame and then selects every i’th unit/person.
• Easy to understand, quick and easy to implement. • Can lead to some stratification depending on how the list
is ordered.• Can be expensive• Need to be careful on how list is ordered to avoid bias.
Systematic SamplingExample
Total units on sampling frame = 60
Want sample of 10
Interval size is 60/10 = 6
Select random start between 1 and 6
Select every sixth unit
1 16 31 462 17 32 473 18 33 484 19 34 495 20 35 506 21 36 517 22 37 528 23 38 539 24 39 54
10 25 40 5511 26 41 5612 27 42 5713 28 43 5814 29 44 5915 30 45 60
Sampling Frame
Stratified Sampling
• Units/people are aggregated into subgroups called strata. A certain number of units are sampled from each stratum.
• Guards against unusual samples• Stratification information has to be available
United Kingdom (60m)
England (51m) Northern Ireland (2m) Scotland (5m) Wales (3m)
Stratified Sampling
Proportionate Stratification• Chance of inclusion in sample is same for all
units/people regardless of strata
Population
United Kingdom 61 000 000
England 51 000 000
Northern Ireland 2 000 000
Scotland 5 000 000
Wales 3 000 000
Sample Size
6 100
5 100
200
500
300
Sampling Fraction
0.01%
0.01%
0.01%
0.01%
0.01%
Stratified Sampling
Disproportionate Stratification• Chance of a unit/person being included in the sample
depends on the strata they are in.• Often used to target small sub groups to help analysis
Population
United Kingdom 61 000 000
England 51 000 000
Northern Ireland 2 000 000
Scotland 5 000 000
Wales 3 000 000
Sample Size
6 100
3 100
1 000
1 000
1 000
Sampling Fraction
0.01%
0.01%
0.05%
0.02%
0.03%
Sampling Frames
• List of all units/people that could be included in sample
• Sample is only as good as the sampling frameo Eligible units/people not on frame cannot be selected
– leads to coverage erroro Units/people on frame more than once changes
probability of being selectedo Ineligible units/people on frame can lead to final
sample being smaller than intended.
Intended and achieved populations
Intended Population
Coverage Bias
Sampling Frame
Sampling Variance
Selected Sample
Non Response
Achieved Sample
Improving Response Rates
Improving Response Rates
• Incentives
• Advanced Letters & Reminders
• Respondent Burden
• Call Back
Incentives
• Incentives – monetary or gift, e.g. pen, gift token?
• Incentives generally improve response rates.
For postal surveys: If the budget is an issue then follow up mailings are preferred, if time is an issue incentives are preferred (Larson and Chow 2003).
– Robertson et al. (2005) found a Lottery scratch card incentive increased postal survey response by 9.6%.
– Prize draws may be ineffective - the incentive is abstract and distant from the participant (McCarty 2006).
Advanced Letters and Reminders
• Cost Effective – they work
• Sponsor – who is the survey for? People are more likely to provide information to the government (or local authority) than to a private company.
• Salience – emphasise the importance a topic may have to a respondent.
…but... Overemphasising importance or sensitivity may put some people off (Groves 2006).
Strike a balance.
Respondent Burden
• Burden –The right questions and only those that need to be asked.
• McCarty et al. (2006) found that a 10 minute increase in survey length results in a 7% decrease in response rates for telephone surveys.
Call Back
• Repeated call back improves response particularly used in face to face interviews.
– Minimum standard: rules about timing and number of call attempts before classification as non-contact. Training techniques to avoid refusal.
– Extended measures: passing on to a senior interviewer; phone call and appointment, letter and appointment; letter from study director.
(Lynn and Clarke 2002)
In postal surveys we can re-issue the questionnaire
General Recommendations
• Include incentives (if the budget is an issue follow ups are preferred)
• Issue advanced letters
• Keep respondent burden as low as possible
• Have a clear call back procedure
• Make it salient
Group Exercise