WEEK 4 L8.1 surveys Flashcards
survey challenges about measurement
process
- measuremet validity
- measurement reliability
concept construct => measurement => collect responses
survey challenges about representation
process
- coverage of pop.
- sampling error (during the selection of respondents- does it introduce a systematic bias)
- non response (are those who don’t respond from a specific group)
target population => sampling frame => sample => responents
types of surveys
- cross sectional: they give you a cross section of the population in one time frame ‘snapshot’ can only allow statistical control
- longtidunial: surveys a population at two or more time periods, and there are 4 versions of that
a. cohort
b. panel
c. rolling cross section
d. trend study
cohort and panel studies are a subcategory of (..) studies
longtidunial
longtidunial cohort study
the same survey is conducted several times each time with a different sample (pooling cross sectional time series)
longtidunial panel survey
when the same participants are questioned at multiple time periods, a cross-lagged causal analysis
- clearly measure changes over time
longtidunial rolling cross section
starting with a very large cross-sectional sample, but instead of interviewing responents at the same time, the interviews are spread out. every day a small subset from the full sample are interviewed.
captures dynamic changes and trends.
disadvantages of rolling cross sectional
lots of time and money required
types of longtidunial surveys
- cohort (pooled cross sectional)
- panel survey (crossed lagged causal analysis)
- rolling cross section (dynamic changes)
- trend study (aggregated time series)
longtidunial trend study
what makes it different?
in the first three types, you still have responses from individual responses, in trend study, you only take aggregate outcomes from seperate polls.
unethical polls
push poll- purpose is persuasion
frugging and sugging- survey is a marketing tool
difference between cohort, and panel
in cohort, a new sample is drawn every time (rs is written), you can’t track individual changes, but you can observe changes at the aggregate level.
in panel, you only do random selection once (the same people are interviewed repeatedly)
age-period-cohort models
tthey can disentangle age effects from period affects, and from cohort effects.
like from young vs. old
critical events that influence everyone
attitudes tied to a specific birth cohort like milennials etc.
which type of survey allows for cross time lagged causal analysis?
longtidunial panels
(1) allows for individual level analysis
(2) allows for conclusions about groups of people
1: panel
2: cohort
how are surveys in terms of internal validity?
Given the correlational design, these tests are not conclusive, and therefore weak in terms of internal validity.
Alternative explanations can only be controlled by adding them as an additional variables to a statistical analysis
Panel surveys are an exception to this.
some possible issues that may arise in questionaires
length shouldnt be too much
it shouldn’t cause reactivity- it should just measure
response scales- neutral points or dont know
question order, and wording
which types of questions should be avoided in questionaires
vague, acronyms, leading, negative, double barred questions, loaded items
some solutions to bias in questionaires?
- randomizing question order and wording
- balanced questions
- pilot study- pre testing
- monitoring and verifying
option with very high/ high costs
interviewer
face to face capi, phone, cati or tde
what’s the cheapest option survey?
self administered
online very cheap, post/mail cheap
what’s the speed of face-to-face capi?
low
what’s the speed of self-admin. post/mail?
low
which survey options are quick?
phone cati or tde
online
what ranks high in terms of control?
face-to-face is very high
phone is high
low for self-admin (post/mail and online)
interviewer effects are high for?
face to face capi, and phone capi, tde (interviewer), and low for self-admin.
how’s coverage for interviewer surveys?
good for face to face, not everyone can be reached by phone (can cause systematic bias- but we can use random digit dialing)
how to deal with selecion bias for phone surveys?
random digit dialing
how’s coverage for post/mail and online?
mail is good
online is biased (systematic bias)
which has the best response rate?
low/ high for face to face