300 likes | 508 Views
Communicating Quantitative Information. Exit polls, Nielsen, Web polls population models Questions for midterm? Homework: Post topic. Postings. Study for midterm. Inflation reprise. Definition?. Inflation. Prices generally go up in steps granularity quantum amounts
E N D
Communicating Quantitative Information Exit polls, Nielsen, Web polls population models Questions for midterm? Homework: Post topic. Postings. Study for midterm
Inflation reprise • Definition?
Inflation • Prices generally go up in steps • granularity • quantum amounts • Movie tickets go up in dollars? • Sometimes, goods and services rise in price by splitting into categories
Tickets, continued • Movie tickets were $6 for everyone. Theater now charges: • daytime: $6 • evening senior citizen $6.50 • evening everyone else $7 • So what is the inflation rate? Compute average ticket. Determine (estimate) amount of tickets in each category • 200 daytime, 50 evening seniors, 400 everyone else (200*6 + 50*6.50 + 400*7) / (200+50+400) is ??? Estimate (guess)
Answer • Estimate: more than 6.50, because overwhelming amount of tickets (400 versus 250) is at $7 • Visualize 3 stacks • Answer: $6.70
Polling procedure • Ask N randomly chosen people (meaning any one in THE population has equal chance of being chosen) • Say p is proportion of sample • Decide confidence level. Use table to get multiplier (z transform). For 95%, use 1.96 • Calculate square root of (p)*(1-p)/N. Multiply this by 1.96. Call this m (margin of error) • Then make statement
Statement 1 • I am 95% sure that the real proportion is between p-m and p+m. • There is a 1/20 chance that the sample was strange (an outlier) and this is not true.
Statement 2 • (Do calculation, but multiply by 2.58, the transform for 99%). Call this number mb. The mb is bigger than m. • I am 99% sure that the real proportion is between p-mb and p+mb. • There is a 1/100 chance that the sample was strange (an outlier) and this is not true.
Calculations? • ? • Yes, you need to know the procedure including the formulas (but you can write them down) for margin of error. • No, you will not be asked to carry out this calculation for sampling on the midterm. • Yes, you need to be able to reason about the formulas. • Yes, there will be some computation • For example, inflation • 2x2 analysis (false positive, etc.)
What if I want to • make the margin of error smaller • need to increase the size of the sample • To halve the margin of error, need to quadruple the size of the sample • Note: you can control sample size. You can't control the answer: • margin of error will decrease as proportion gets farther from .5 so depending on the result, there will be a different margin of error.
How • to get samples • Note: sample quality is important. • Sample blunders lead to bad predictions!
Bradley effect • 1982. Tom Bradley, African-American, lost the race for governor of California after being ahead in the polls. • Also, exit polls had him ahead • Recent update from person in the campaign • He wasn't ahead. • Internal poll tied. • Others, ahead but within margin of error • He did win on election day, but not in absentee ballots which were enough to make a difference.
2008 • Would there be a Bradley effect? Was Obama ahead in the polls because people didn't want to admit they wouldn't vote for a "skinny guy with a funny name", that is, an African-American…. • Answer: there wasn't a Bradley effect for Bradley! But still an issue of what people tell pollsters.
Opportunity sample • Common situation • people assigned/asked to have a meter attached to their TVs • people asked/voluntarily sign up to have a meter (software) installed in their computers. • Practice is to determine categories (demographics) and project the sample results to the subpopulation to the population • Has negative features of any opportunity sample • Are these folks different than others in their (sub)population?
Requirements • Model / Categories must be well-defined and valid • Hispanic versus (Cuban, others) in Florida in 2000 • Need independent analysis of subpopulations representation in general population • The sample sizes are the individual Ns, making the margin of errors larger
Standard error is the success proportion in sample of size n
Sample as set of samples • Let's assume task is to report on visits to Web sites. • Recruit sample: turns out that actual sample is 60% men and 40 % women out of 100 subjects.men in sample visited women visited google 10000 8000 ebay 20000 6000 • Actual population (determined by some other means, maybe phone polling) is(absolute numbers) 27000000 men 25000000 • To get numbers for whole population, 27000000/60 = M/10000 M = 10000*27000000/6025000000/40 = W/8000 W = 8000*25000000/40 M + W is estimate of visits by population to google
Election polling • People polled versus 'likely voter' • Some politician: What do you call a candidate who depends on new voters? A loser. • Exit polls do identify voters, but • some may refuse • some may lie • early voters may differ from late-in-the day voters • Election polls have been accurate • within margins. Recent elections have been very close! Polls correctly identified 'swing states'. Nate Silver (538) got 52/53 (DC, Nebraska) correct.
Past event • 1936 Liberty poll • based on mailing postcards to people with phones and people with cars • Problem? • response rate was okay • Problem? • Predicted Wilkie victory (FDR loss).
Past events • Kinsey interviews • Shere Hite • ? • Note: when there is little or no base information, any findings can be valuable • Qualitative versus quantitative studies
Past event • According to Cartoon book (published in 1993): • "Last 5 presidential elections….Gallop poll interviewed fewer than 4000 voters. Each time errors in predicting presidential election outcome less than 2%."Note:outcome, not just winnerNote: timing of predictions is not mentioned. • Reminder: 1992: Clinton beat Bush 1988: Bush beat Dukakis 1984: Reagan beat Mondale 1980: Reagan beat Carter 1976: Carter beat Ford • Model, 'bias corrector', process more important than sample size (over a few thousand).
Exact form of question • … often doesn't get included in news story • Belief in Holocaust survey: news story appeared that 30-40% of people in USA do not believe there was a Holocaust. • refuted by second researcher in two ways • same format, question about moon walks • different (better) question • Good project topic • Supporting abortion surveys
Original question:"Does it seem possible or does it seem impossible to you that the Nazi extermination of the Jews never happened?" • Recommissioned poll:"Does it seem possible to you that the Nazi extermination of the Jews never happened, or do you feel certain that it happened?"
Polling topic • Recall themes • definitions • sample, N, margin of error, confidence • denominator • What's the difference? comparison / context • qualitative aspect: sample representing which population • Add new themes • statistical technique • assumptions regarding distributions
Reports on polling • What was • asked versus • the headline versus • featured in the story versus • in the story • (or is) your idea/understanding of issue • Who • was asked (how was the sample made) vs • population described in story vs • implicit idea of population • When • Where
What was asked • Regarding 'war on terror'http://www.nytimes.com/2007/10/17/us/politics/17web-elder.html?_r=1&8dpc&oref=slogin • Push polling: • Asking questions, but really trying to 'plant the seed' for certain ideas. • Operational definition(s) are necessary
Zogby polls • I signed up to do this. This is a ‘panel’ form of voting: self-selection. Questions are asked to categorize responders. • "NASCAR fan", "member of investor class", plus regular demographics • A 2006 poll was (IMHO) potentially misleading • Question: will Foley incident have effect on my vote • Votes versus non-votes • Did [some / many] Republicans stay home? • Need to see how this is reported: • ‘what is universe’ question relates to ‘what is denominator’ question
Similarly, studies… • such as value of low-fat diet, calcium, arthritis drugs have issues of • What • low fat versus low specific fats • intent versus compliance • this was an issue in the low fat diet AND even the calcium plus placebo diet • ‘retrospective’ study (People recalling what they did.) • Who • results for specific populations
Homework • Topic for project 1 due by October 21 • Monday is review day for midterm • Read and study the guide • Midterm is October 18 • VOTE Nov. 2nd. • Extra credit: take picture of yourself and email it or bring to class. Other proof? • Presentation/paper project 1 paper due Nov. 4 • Continue postings • originals and responses