1Probability Proportionate to Size• A type of cluster sampling where a cluster's probability of being selected is proportional to its size. That is, the larger a cluster, the higher its probability of being selected. • Within each cluster, a fixed number of cases is selected. For element i in cluster j (by the rule of conditional probability):• Equal probability for each elementP(ij) = P(i | j)*P( j) =1nj⋅njn=1nNonprobability Sampling1. Convenience sampling• Reliance on available subjects.• E.g., sample friends, coworkers2. Purposive sampling• Selecting a sample for a purpose, usually because of their unique positions.• e.g., sample state officials to learn about welfare reform.Nonprobability Sampling3. Quota sampling• Begins with a matrix of the target sample.• Data are collected from people with the characteristics of a given cell. • Compare with stratified sampling.• Problem: bias in other characteristics; must know population characteristics to set the quotas or weights.2Nonprobability Sampling4. Snowball sampling (respondent-driven sampling)• Researcher asks respondents to recruit peers. • Useful when members of a population are difficult to locate (homeless, migrant workers, prostitutes).• Useful for study networks, not for isolates (e.g. murderers).• For snowball sampling to work well, use more seeds, longer chains.Review: Sampling Error• We are interested in certain characteristics of a population. Such characteristics are called population parameters. • We do not observe the whole population. Instead, we draw a sample and calculate sample statistics. We use sample statistics as estimates of population parameters.• Sampling error is differences between population and the sample that are due only to chance.Sampling Design and Sampling Error• Stratified sampling – Reduces sampling error (more efficient than SRS)– Requires prior knowledge of population proportions– Most advantageous if strata are externally heterogeneous.• Cluster sampling – Save transportation costs– Increases sampling error (less efficient than SRS)– because clusters tend to be externally heterogeneous and internally homogeneous.3Sources of Error• Sampling error (variability)– Sampling errors diminish as sample size increases. • Non-sampling error (bias)– Non-sampling errors are called biases, which do not diminish as sample size increases.– Errors of observation (poor measurement) – Errors of non-observation (omission of cases that should have been included in the sample)– Inadequate coverage of the population (poor sampling frame)Class 16Survey ResearchSurvey Research• Survey Modes• Exercise• Example: “AddHealth” Research Design• Guidelines on Asking Questions4Modes of Survey• Self-administered mail survey• Face-to-face interview survey • Telephone interview survey• Web survey• Multi-mode surveySelf-Administered Mail Survey • Steps1. Notification letter2. Mail distribution and return3. Monitoring returns4. Follow-up mailings (reminder postcard, replacement questionnaire)• Advantages– Relatively quick– Low cost– Can offer anonymitySelf-Administered Mail Survey• Disadvantages– Requires literacy– Low response rates (non-response biases)– Cannot use complex skipping patterns– Responses may not be independent– Interviewer cannot observe5Face-to-face Interview Survey• Steps1. Notification letter2. Confidentiality form3. CAPI (computer-assisted personal interviewing) – Easy to develop skip patterns– Easy to use split ballot designAlso applies to CATI and web surveyFace-to-face Interview Survey• Advantages – Fewer incomplete questionnaires– Higher response rate– Interviewer can take observation notes– Appropriate for complex questionnaires– Allows for relatively long questionnaires• Disadvantages –Higher cost – Slow field work– Interview effects and interviewer effectsTelephone Interview Survey• CATI (computer-assisted telephone interview)• Advantages– Lower cost than FTF interview surveys– Quick implementation (e.g. political polls)– Easier to supervise field work • Disadvantages – Questions must be short and simple– Cannot use visual aids– Potential biases due to selectivity of a sampling frame (phonebook & random digit dialing)6Web Survey• Advantages– Low cost– Quick implementation– Data entry errors are eliminated– Quick results• Disadvantage– Selective sample– Not good at ensuring completionImproving the Response Rate• Give an incentive (gift, cash).• Send a notification letter.• Follow up on non-response.– Mail survey and web survey: send a reminder; make questionnaire attractive.– Face-to-face or telephone survey: Vary the days and times to visit/call. Comparison of Various Survey ModesGoodPoorGoodPoorTime-savingGoodPoorAverageGoodLower costGoodPoorAverageGoodAvoid distortion of answersAverageGoodAverageAverageAllowable length of questionnaireGoodGoodPoorAverageHandling complex questionsPoorGoodAverageAverageQuestionnaire completionPoorGoodGoodAverageResponse ratePoorGoodGoodAverageReaching a representative sampleWebF-T-FPhoneMail7Secondary Data Analysis• Secondary data analysis has become the primary means of conducting empirical research. • Reasons:– Concern with generalizability.– Increased cost of conducting independent surveys.– Wide availability of existing data, such as GSS.The editor of the Wisconsin State Journal has asked for proposals to conduct a “final” election survey prior to the general election, to be held on Nov. 7, 2006. In the past, this newspaper has relied on telephone surveys, as have most other newspapers in the U.S. Typically, 800-1200 households are polled in a 3-4 day period prior to the Sunday before Election Day. The contractor must provide results of its poll, including estimates of the percent of Wisconsin voters who will vote for each state and congressional representative, governor, and all state-elected officials, plus any initiatives that appear on that year’s ballot. In addition, several demographic questions (age, education, party preference, income, gender) will be asked to help interpret data. The final data set must arrive at the Wisconsin State Journal offices by 1 PM Sunday November 4. The WSJ has received three proposals with similar budget estimates. Witch one would you recommend be accepted and why? Proposal 1 A telephone poll, 10-12 minutes long,
View Full Document