Stats

[Thomas154321]

1. The author Shere Hite undertook a study of American women's attitudes towards sex and love by distributing 100,000 questionnaires through women's groups. Only 4.5% of questionnaires were returned. Based on this sample of women, Hite wrote Women and Love, a best-selling book claiming that women are fed up with men. For example, 91% of the divorced women in the sample claimed that they had initiated the divorce, and 70% of the married women said that they had committed adultery.
   
   Explain briefly why Hite's sampling method is nearly certain to produce a strong bias. Are the sample results cited (91% and 70%) likely to be much higher or much lower than the truth aout the population of all adult American women?
   
   
   
   
   
   
2. According to an observation study carried out at Kaiser Permanente in walnut Creek, California, users of oral contraceptives have a higher rate of cervical cancer than nonusers, even after adjusting for age, education, marital status. religion, and smoking. Investigators concluded that the pill causes cervical cancer. Discuss briefly whether they were right to do so.
   
   
   
   
   
   
3. The data in the table below are extracted from the Which? guide to new and used cars 1994, a publication of the Consumer's Association. 66500 randomly selected Which? magazine readers received questionnaires concerning car reliability and so forth. Of this population, 26340 readers replied, and the third column in the table shows, for example, that 1096 BMW owners responded. The second column shows the approximate market share of each mannufacturer.
   
   To which population do you think the summaries are intended to refer? To which population do the summaries actually refer> Are these the same? How representative is the sample? [Hint: draw a graph of market share versus sample size, and compare what you see with what common sense leads you to expect.]
   
   

Not really sure at all about these. For 1 I think the sample is far too small and also those with bad experiences are far more likely to write in and complain. For 2, I think it is wrong because correlation does not imply cause, for example just because ice cream sales and beach towel sales both increase during summer, you cannot say that ice cream sales cause beach towel sales or vice versa, although that seems like quite a weak argument to me... For 3, a) The statistics are intended to reflect actual market share in North America b) However the statistics actually are representative of Which readers and therefore NOT necessarily the same as the actual market share values. Not sure what graph it wants or what it expects though, any help there would be appreciated.



Thanks for any help offered.

[zaza]

1) A sample of 4500 women, all of whom attend "women's groups" and all of whom had the choice to participate, leads to conclusions being drawn about women in general?

2) Also, what about implied effects? Suppose it is unprotected sex that leads to a higher rate of cervical cancer?

3) The statistics compare market share amongst the population with samples from Which readers. Note that Volkswagen and Rover have similar sample sizes, but wildly different market shares. If the stats were representative of the population, you would expect a graph of market share vs sample size to correlate, would you not. Draw the graph and see...


zaza