Week 11 Flashcards
Data collection method - Personal interviewing
4 pros
2 cons
Pros
1. Better cooperation and rapport
2. Can probe for adequate answers, help complex instructions to be followed
3. Can combine with providing visual cues and with other computer-based modes
4. Better at longer questionnaires
Cons
1. More costly
- need to train interviewers & send them to diff. locations
2. May require more travel
Data collection method - Telephone
4 pros
3 cons + 1
Pros
1. Cheaper than personal interviewing
2. Can access some populations better than personal visits
3. Interviews may be shorter
4. Easier interview staffing and management
- {no need to arrange travel schedules}
Cons
1. May have sampling problems, eg. coverage (ppl w/o phones)
2. More nonresponse than personal interviewing
- ppl can just hang up
3. Cannot use visual aids
+ Need to be very short, since respondents tire quickly and may hang up
Data collection method - Mail
4 pros
3 cons + 2
Pros
1. Lower costs
2. Require less staffing
3. Can access some kinds of ppl more easily
4. Respondent has more time to provide thoughtful answers
Cons
1. Often higher nonresponse
2. Lack of interviewer to assist respondent
- but also no interviewer effect
3. Accurate mail address may not be available
+ Give no control over who answers (e.g. survey sent to man, wife responds)
+ Require simple structure since no elaboration is possible
Data collection method - Internet
5 pros
3 cons + 1
Pros
1. Low costs, once system set up
2. Potential high speed of return
+ advantages of measurement with self-administered instrument, like mail:
3. Require less staffing
4. Can access some kinds of ppl more easily
5. Respondent has more time to provide thoughtful answers
Cons
1. Restricted to population of Internet users
2. May lack sampling frames (not clear for Internet)
3. May have high nonresponse
- very many online surveys exist
+ Are likely to be seriously biased (are respondents representative?)
Nonresponse
The failure to obtain complete responses to the survey questions on eligible units in the selected sample.
Unit nonresponse vs Item nonresponse
Unit nonresponse - failure to obtain responses to ANY survey questions
Item nonresponse - failure to obtain responses to a SUBSET of the questions
Undercoverage
Units in the target population which are not in the sampled study population,
i.e. have no chance of being selected into the sample // not covered by sampling frame
eg. people w/o phones when the sample is selected using telephone numbers.
4 sources of unit nonresponse
- Non-contact
- failure to locate the sample unit or to contact the unit. - Non-cooperation
- refusal of sampling unit to respond - Inability to respond
eg. language barriers, ill health - Other, eg. Accidental loss of data/questionnaire
3 main sources of item nonresponse
- Respondent
- did not know answer
- refusal (sensitive/irrelevant question)
- accidental skip (can be avoided by computer-based Qs) - Interviewer
- did not ask Q
- did not record response - Processing
- response rejected at editing
Nonresponse bias
2 formulae
Systematic bias resulting from systematic differences between respondents & non-respondents
- reduces sample size, therefore increases std errors & decreases precision
^should select larger initial sample
Bias = E(yR bar) - yU bar = yRU bar - yU bar
Bias = (1-NR/N) (yRU bar - yMU bar) where NR/N is the response rate
4 ways to prevent nonresponse during data collection
- Minimise noncontact
- Callbacks on a careful schedule may reduce interview non-response - Minimise refusal
- Rewards and incentives (e.g. M&S vouchers)
- Use of carefully trained interviewers - Allow proxy respondents
- Substitute non-respondents (not generally recommended)
*can also deal with nonresponse through post-hoc stratification
Cohort study (from ex.9)
- An analytical study which tries to establish cause and effect using observational data
- A group of people (the cohort) is followed over time, to examine associations between different interventions received and subsequent outcomes.
- The different interventions (vaccine vs no vaccine) are compared with each other.
Explain briefly what is meant by a PRESTIGE ERROR. Include two examples in your explanation.
[4m, 2019]
Prestige errors arise when respondents feel their social status is under threat. They may claim to brush their teeth more than they do, read more books than they do, buy new clothes when they use charity shops etc.
Why is non-response often a serious problem? [3m, 2019]
- Non-respondents will REDUCE the SIZE of a survey. {smaller sample size, larger variance}
- More seriously, those who don’t take part may be DIFFERENT from those who do (on the variable of interest),
- and this will lead directly to a BIASED ESTIMATE.
Response rate
[2014]
The proportion of eligible sample units that successfully provide data
{=no. of respondents / sample size}
