Ch7

sampling - process of selecting observations
probability sampling - generalizable from sample to population, requires random selection.

NONprobability Sampling:

• Reliance on Available Subjects/Convenience - no control over representativeness of sample, generalization from data is V. limited.

• Purposive/Judgmental - goal to compare left-wing and r-wing students, so you sample 2 orgs only - your goal is comparison so this is OK, but can't generalize to L and R-wing students in general. Also, interviewing students who don't attend school rally to learn about school spirit.

• Snowball - when pop difficult to locate (homeless introduce to other homeless, who introduce to other homeless...). Results are of questionable representativeness, used freq for exploratory res.

• Quota sampling - sampling based on knowledge of a population's characteristics. Selection of sample to match set of characteristics. Quotas based on vars most relevant to study. Quota frame MUST be accurate. (what is QF?)

• Selecting Informants - member of group who can talk about group. Be V careful with selection, to pick a most normal, centrally accessible, many contacts informant.

Probability Sampling:

• to provide useful descriptions of total pop, sample must contain same variations that exist in pop. Every member of pop has = chance of being selected for sample.

• Bias: selections are not typical or representative of larger pop.

• Sample is rep of pop if aggregate char of sample approximate same aggregate chars in pop. Samples that have equal probability of selection method samples are labeled EPSEM.

Element - unit info is collected on and provides basis of analysis (generally people)
Population - “theoretically specified aggregation of study elements” (dumb)
Study Population - aggregation of elements from which sample is actually selected. (Can be artificially limited).
Random Selection - each element has = chance of selection independent of other event in selection process. ('selection' of head or tail in quarter flipping is independent of previous coin tosses)
Sampling Unit - element or set of elements considered for selection in some stage of sampling (??? How is this different from study pop? )
Probability Theory - sampling teqs that produce representative samples and analyze results of sampling statistically. Provides basis for estimating parameters of a pop.
Parameter - summary description of a var in a POPULATION.
Statistic - summary description of a var in a SAMPLE
Sampling error - degree of error to be expected for given sample design (allowed bec of probability theory -- how closely are sample stats clustered around true value?)

What do we have to know of sampling and std error? I've covered nothing, here.

Confidence interval/level of confidence - express accuracy of sample stats in terms of level of confidence that stats fall w/in an interval from the parameter (sample estimate, cause we don't know the Parameter). Additionally, provides basis for det sample size.

How much of this do we need to know?

Sampling frame - list of elements from which probability sample is selected. (sample of students taken from roster, then the roster is the sampling frame). Why is this not the study pop?!!

Types of Sampling Designs:

Simple Random - once sampling frame established, assign random number to element in list, then use table of random numbers to pick study sample.

Systematic - every kth element in a sampling frame, with first element sel at random. Be V AWARE of the dangers of periodicity (every 5h house on block is the corner house, and as such, is abnormal).
Sampling interval - std dist bet elements sel in sample.
Sampling ratio is proportion of elements in pop that are selected. (1/10 if every 10^th person is selected)

Stratified - greater degree of representativeness. Org list into homogeneous subsets and pull every kth element from that list, making sure that subsets are in same proportion as pop.

Multistage Cluster Sampling
: initial sampling of clusters, then sel of elements w/in each sel cluster.

- Highly efficient but less accurate sample. 2 stage cluster sample is subject to two sampling errors. Maximize # of clusters while dec # of elements w/in each cluster.

• Typically elements composing given cluster w/in pop are more homogeneous than is total pop. (residents of block more alike than nation)

• Additionally Multistage cluster sampling can be Stratified.

PPS (Probability Proportionate to Size) Sampling: type of cluster sampling. Used when clusters are of greatly differing sizes (ie city block vs suburban block) so that everyone gets the same = chance of being selected.

Further refinement is the Disproportionate Sample and Weighing: Disproportionate - you may decide to sample to get higher # of some small subpopulation so that you can have sufficient #s to analyze results with some meaning. Analysis of 2 samples needs to be separate - but then they can be compared. Weighting happens when you want a composite sample of the entire pop, then you have to weigh samples when 'adding them back together'.