Lecture 05
SOCI 313
July 17 , 2013
Population and Sample
-whom and what will we collect data from
-describe in terms of variation
-knowing about a population, takes a lot of effort to capture variation, really dependson how
much variation is in it
-if the population is homogneous, then the process can be incredibly simple
-develop a sample where we will capture variation
-than generalized back to general population, estimate the error in doing that
-fundamental problem behind sampling is to find the same variations that occur in the population
-no way we generalize without it
=in orer for us to properly respresent what is going in the population, have to make sure there is
no error
-biggest source of error is some form of human bias, randomize your sampling, leave the
decisions to chance
-preferneces bias your ability to go out and collect data
-understnading of population there fore differe, mustdo this by removing human decision making
by using randomization to rule it out
-bias not only affects our ability go generalize to the larger population also effects our ability to
make the relationship between x and y
Sampling frames and random sampling
-bias of human erroridn’t have a large enough sample, general rule is at minum 30 people in a
sample because you can be assured that random selection has actually occurred
-assumption in smaller populaitons, refence to large population
-need to have certain number of elemnts drawn into your smaple in order to make it accurately
represent the population
Sampling Error
-simple random sampling is theoretically the best way to create a represetntative sample of a
population, because it minimizes sampling error
-random sampling minimizes theoretical sampling error
-systematic error is a human bias
-is a bias in your sample, obviously unrepresentative of the variation in the population attribute
that has occurred because of other reasons
-more likely that some elements will be systematically drawn than others
-best way to remove random error is to increase sample size, really comes down to in absolute
terms an continue to perfect terms by increasing sample size, but there is a limit determined by
a payout
-diminishing returns in terms of human labout, very minimal improvements when looking at
extremely large samples
-national polls only use about 10,000 people, only want to know that estimate for the population
are precise
-but for sociology for generating theory, a 1-2G is just as good as 10G
-for theory generation -do not need accuracy, and do not have means and money to do so either
-random pull from pull into sample population is randomly designed, by every other derivation si
designed
-systematic sample with a random start
periodicity human error
-oeridoicitiy occurs because elements are lined up due to eg gender and racialized status
-sampling frame sorting also impacts bias, can also create a bias
-look at males vs females eg for stratified sample
-the researcher first divides the population into grouping or strata of interest
-then samples from a sampling frame
How to overcome random stratification: using a census instead of a sample
-one to one ratio, for a smaller population
-disproporational sample, can properly represent a smaller group in comparison to larger ones
-although lose the abi

More
Less