1 Social Research Methods Surveys
2 Survey Characteristics Collecting a SMALL amount of data in STANDARDISED form from RELATIVELY LARGE NUMBERS OF INDIVIDUALS Selection of REPRESENTATIVE SAMPLES of individuals from KNOWN POPULATIONS
3 Surveys are very common Most reported research is a survey Most calls for research want surveys done.
4 References de Vaus, D. A. (2002) Surveys in Social Research 5th ed. London: Routledge. Aldridge, Alan, Levine, Kenneth (2001) Surveying the social world : principles and practice in survey research. Buckingham; Philadelphia, PA: Open University Press. Fowler, Floyd J. (2008) Survey research methods 4th ed. London: Sage Publications. Hoinville, G., Jowell, R and associates (1985) Survey Research Practice. London: Gower Moser, C. A. and Kalton, G (1971) Survey Methods in Social Investigation. Aldershot: Gower
5 Mounting a survey Initial design including sample design and selection Questionnaire construction (incl. piloting) Fieldwork (the most efficient stage) How to deliver? Postal, self-administered, handout, , WWW etc. In person/interview/telephone Dealing with non-responses Editing and Coding Computer entry and editing Analysis and interpretation Write Up.
6 Remote vs. Interviews ProCon Often only/easiest way to get info. Low response rate, may not be representative efficient, low cost, fast Ambiguities not detected anonymousRespondents don’t treat seriously Remote (= postal, www etc.)
7 Interview ProCon Can clarify questionsInterviewer bias Interviewer encourages participation/ involvement Interviewer-respondent effects May not be seen as anonymous. Respondents less honest
8 Survey design 1. Simple Survey e.g. ad hoc sample survey (a snapshot), cross sectional survey Essentially descriptive - says what the current state of affairs is. 2. Panel Survey Collect information from same people at 2 or more points in time. 3. Rotating sample survey Repeated survey with some continuing respondents and some new. 4. Longitudinal survey either a form of panel over a long period of time or simulated by combining simple surveys from different time periods.
9 Sampling Why sample? 1. Population too large to take all e.g. UK population 2. Population too dispersed or difficult to contact e.g. members of swimming clubs 3. To get results quickly Key to sampling = getting a REPRESENTATIVE sample (compare with blood sample)
10 How big a sample? 1. Less than 1/10 of the population 2. Big enough to produce acceptable sampling error (e.g. about 2,000 from large population) 3. Big enough to give reasonable numbers in subsets (e.g. 4 x 5 table = 20 cells, need 10 in each, so sample ≥ Small enough to carry out with available resources and time.
11 Types of sampling A.Probability each member of population has known and usually equal chance of inclusion. Can make statistical inferences B.Non-probability select by non-random methods involving human judgement. Sample may be representative but this not known
12 Most probability sampling needs a sampling frame A list of everyone in the population + means of contacting. N.B. problems 1. Categories excluded (e.g. all students misses occasional students) 2. Inaccuracies (e.g. in when should be off list, off when should be in) 3. Sampling frame in wrong units (voters when household needed) 4. No information on population which is needed for stratification 5. No means of contact (e.g. list of church members, no addresses)
13 Types of probability sample Simple random Systematic or quasi-random Stratified Multi-stage Cluster
14 Simple random Every member has equal chance of selection Called epsem design (Equal Probability Selection Method) Select from sampling frame by: a.lottery (e.g. toss coins) b.using list of random numbers c.computer generated random number list
15 Systematic or quasi-random Simple random sampling very inconvenient for large samples Therefore Start from random point and then take every r th person on list e.g. 5% sample from population of 2,000 (=N) i.e. sample of 100 (=n) r = N/n = 2,000/100 = 20 Therefore select every 20 th person Start from random person between 1 and 20, say 13 so take person 13, person 33, 53, 73 etc. up to 1,993 N.B. only 20 possible samples compared with billion of billions for simple random sampling
16 Stratified Used to increase likelihood of representative sample (i.e. reduces sampling error) Divide population into strata and ensure share of sample is from each stratum. e.g. age groups. if 13.5% of population aged then 13.5% of sample should be aged Needs information on stratification factor in sampling frame. Strata should be internally homogeneous and different from other strata. Stratification factor should be related to issues of interest in survey
17 Disproportionate stratification Increase chance of selecting from some strata Adjust for this in analysis
18 Multi-stage Increases sampling error but decreases cost of sample Increases geographical concentration Divide sampling frame into groups and select only some groups (e.g. counties, towns) May be several stages Sampling with probability proportional to size Adjust chance of selection within group to account for different sizes of groups
19 Cluster Used where no final sampling frame available Like multi-stage, select some groups, at several levels then take ALL people in selected groups. Like multi-stage it concentrates sample but increases sampling error.
20 Non-probability sampling Not chosen using probability methods - rather some human judgement Most common = QUOTA sampling Samples are stratified usually by age, sex, occupational status
21 QUOTA sampling People chosen by interviewer (not at random) from those passing by to fill quotas of e.g. so many men, so many women, so many aged etc. Quotas can be linked e.g. so many men aged N.B. no estimate of sampling error possible BUT evidence suggests good quota samples are good representative samples
22 Other kinds of non-random sampling 1.Convenience sample Friends, neighbours relations etc. 2.Snowball (or cascade) sample Friends of friends, contacts of contacts etc. 3.Purposive Collected for specific purpose or reason