Flashcards

GEG306 - Chapter 7 Part 2

Bob Harold English 35 cards

Cards in this deck (35)

1Probability Theory 1
2Probability Theory 2
3Probability Theory 3
4Probability Theory 4
5Probability Theory 5
6Probability Theory Normal Distribution
7Confidence Intervals for a Percentage
8Confidence Intervals for a Percentage 2
9Confidence Levelusually 95% or 99% (the percentage of time that our confidence interval is correct)
10Confidence Intervalthe range in which the population parameter lies. Also known as the MOE, or Margin of Error.
We make statements like: We can be 95% confident that the population percentage like between 45% and 55%.
11Margin of Error by Sample Size (95% confidence interval 50-50 split)
12Sample Size Small PopulationsNote then that sample size has little to do with the size of the population from which you are sampling
Sample size we need for all the USA is the same as the sample size we need for all of China, even though China has five times the population
If sampling 400 from, say 2,200 faculty, the sampling ratio or sampling fraction will be 400/2200 or 18.1%
For 95% confidence level of a ±5% confidence interval, divide the 400 by 1.181 to yield a sample size of 339.
13Sample Size Blanket SampleIf sampling from a small population, it might make sense to try to get the entire population to respond (all parents who send their children to a particular school
14Sample Size Determinants of sample size includeCost
Time
Geography
Desired Level of Accuracy
Subgroup analysis
Number of age groups or income groups desired
(When you “draw” the sample, must oversample to allow for cooperation rate)
15Sample Size Fixed Costdo not vary with sample size (survey design time from professional, data analysis, report writing) programming questionnaire in CATI
16Sample Size Variable CostInterviewer time, printing, postage, cost of more sampling
Cost of coding responses to open-ended questions
CATI – interviewer time only, for the most part
17Sample Size TimeCan get a large sample size in a small time period by hiring more labor
18Sample Size GeographyIf covering a larger geographic area one can expect greater demographic and other variations, particularly if there is a significant change in environment from one part of the study area to another
19Sample Size –Subgroup AnalysisGiven the sample size determined by other factors, will you have a large enough sample of:
Women faculty
Faculty in a particular school
Full Professors, etc.
20Sampling Framesare lists of all elements in the target population:
List of Households
Lists of Persons
Lists of Companies
Maps
Time periods during which events occur
Lists may not be available
21Sampling Frames ElementsElements are fundamental units of populations:
Households
Persons
Companies
Crime victims
22Sampling Frames Target Populationis the group of elements for which the survey investigator wants to make inferences
23Coverage Properties of Sampling Frames CoveredHow well does the sampling frame (the available materials for sample selection) actually cover the target population.
A sampling frame is perfect when there is a one-to-one mapping of frame elements to target population elements. In practice: does not exist!
Covered (good)
24Coverage Properties of Sampling Frames Under coverageGated communities and locked buildings
Housing in alleys and not visible from street
Groups quarters in some cultures may make it difficult to determine households
25Common Sampling FramesMaps, air photos, and satellite images
CDS (Computerized Delivery Sequence) file provided by the USPS
Telephone or household directories
Lists of email addresses
More than one person can share an email address
Some people do not have an email address
26Clustering of Target Population Elements within FrameInterviewing more than one person in a households is almost never done
If you sample one person they may talk to the second person about the interview before you do the sampling and that will impact their answers
27Sample Design – Respondent SelectionIf do not use, can either pretend that respondents are a random sample of persons, or ask questions such as marital status of all adults in the household
28Multiple Sampling FramesUse a principal frame that covers most
Add a supplemental frame that provide better coverage for those omitted from principal frame
29Probability Sampling Simple Random Sampling (SRS) EPSEMEvery element in the population has a known and equal chance to be in the sample
If select the same element twice, we do not interview that person twice
It is an “EPSEM” sample: Equal Probability Selection Method
30Probability Sampling Systematic SamplingUse when difficult to number all items in a sampling frame (on a printed listing)
You are basically taking every kth element in the sampling frame, where k is the sampling interval
If are sampling 100 out of 1,000 elements, sampling interval would be k = 10
Use a random start, and then sample every tenth
If need 400 out of 5,000, sampling interval is 12.5, perhaps sample every 13th and then pick a few more via SRS (or every 12th and then use a SRS to delete a few)
Use of several random starts (systematic sampling with a random start)
Column inch with a ruler on a long printed list
Problem of periodicity – every 6th housing unit yields a corner condominium, sampling ratio is every 12th month yields data from June of every year (impacts on number of births data)
31Probability Sampling Stratified Sampling - Proportionate
32Probability Sampling Stratified Sampling
33Stratified Sampling - DisproportionateIf over sample some group is called a “disproportionate stratified sample”
If 80% live in Region A and 20% in Region B and sample size is 500, will get 400 in A and 100 in B. May want to do 250 in each area
Example – Getting more households with children in South Palm Beach
If more expensive in one area, may want to under sample that area
34Multi-Stage Cluster SamplingCluster sample (sample size needs to be about 1.5 times as big for same level of accuracy)
Cannot get a sampling frame of all elements, but can get a sampling frame of clusters of elements
Each cluster is called a PSU, or Primary Sampling Unit
University students, church members, gangs
May not be able to get a list of all people in a city, so use a multi-stage design
Sample city blocks
List households on the sampled blocks
Sample the households
List the people in each household
Sample the people in each selected household
Cluster sampling used in home interview survey for reasons of logistics
35—