Term
|
Definition
| The percentage of the civilian labor force that is unemployed. |
|
|
Term
| What is the Bureau of Labor Statistics? |
|
Definition
| The agency in charge of the employment studies and statistics, |
|
|
Term
| For the Current Population Survey monthly report, how many are interviewed by the field staff of the Bureau of Labor Statistics? |
|
Definition
| 1,700 staff members interview close to 110,000 people ( roughly 1/10th of a million ) |
|
|
Term
| In the 1990 redesign of the Current Population Survey how many Primary Sampling Units (PSU) were there? |
|
Definition
| There were 2,007 Primary Sampling Units, each being a city or a county. |
|
|
Term
| Why are the interviews restricted to the same 792 PSU’s out of the 2,007 first made? |
|
Definition
So that the PSU’s in each of the 792 strata will resemble each other in characteristics:
-like number of male and female heads of household
-number of workers in retail trade, etc. |
|
|
Term
|
Definition
| Each of 792 PSUs was divided into Ultimate Sampling Units. A USU consists of 4 housing units each. |
|
|
Term
| At the second stage, what are selected at random? |
|
Definition
| Some USU’s are selected at random and every person age 16 and over are added to the sample for the Current Population Survey |
|
|
Term
| Following stage one and two, how many samples are made in the early 1990s and why that many different samples are made? |
|
Definition
| 16 different samples are made so that rotation of part of the sample can be possible. After put into the sample, a housing unit is kept for 4 months, dropped out dropped out for the next 8 months, and again brought back for final 4 months. |
|
|
Term
| What is the threshold size of a sub-sample allowed in estimation of numbers in Employment and Earnings? |
|
Definition
| A thousand estimates for numbers in Employment and Earnings are done and when a sub sample size is below 50, no estimate is undertaken. |
|
|
Term
| Why not change the housing unit every month in stead of keeping it for 4 months and bringing back for final 4 months with a break for 8 months? |
|
Definition
| That way a lot of money is saved. Also the overlap therein helps estimating the monthly change in Employment and Unemployment. |
|
|
Term
| Why is calculating the SE’s for the Bureau’s estimates a delicate business? |
|
Definition
| Due to the clusters in the sample. Each USU is a cluster. The simple random sample formulas ( with SD of Box etc ) do not apply to cluster samples. |
|
|
Term
| Which has more information : cluster sample or simple random sample? Why? |
|
Definition
| When size is same for both a simple random sample and a cluster sample, a simple random sample has more information. This is because in simple random sample, a person and his neighbor have remote chance of getting into the sample unlike in the cluster sample. A dad and son in a USU will give same information to the investigator. |
|
|
Term
| What is half-sample method? Why is it needed in the Current Population Survey sample? |
|
Definition
| A second sample will cost $40 million a year. So in stead of getting 2 samples, the Bureau splits the survey into 2 independent pieces that have the same chance behavior. Then takes the difference of the estimates and this serves as the SE. The pooled estimate of the two ( average of the two ) is taken as the estimate of the characteristic to be estimated. |
|
|
Term
| In the Survey which has a more accurate estimate :size of the labor force or number of unemployed as compared to simple random sample ? |
|
Definition
Bureau’s SE for estimating the size of the labor force is 5% smaller than for a simple random sample, because weights are doing a good job.
Bureau’s SE for estimating the size of the unemployment is 50% worse than for a simple random sample, because the clustering hurts. |
|
|
Term
| What is the whole lesson from the design of the Current Population Survey sample? |
|
Definition
| To compute a standard error properly, you need more than the sample data; you need to know how the sample was picked. Recall that the sample for rhe Bereau’s work is done with clusters. |
|
|
Term
| Why does the bureau estimate the population percentage of the labor force with different weights to groups? |
|
Definition
| The sampling rate is different from stratum to stratum. If any group is over-represented in the sample than in the population, smaller weights are given to bring the sample back into line with the population. Same way for under-representation, a little increase in weight is made. Adjusting the weights helps correcting imbalances caused by chance variation and reducing sampling error. |
|
|
Term
| Why survey data are considered as more accurate than census data? |
|
Definition
| Unlike census the Current Population Surveys operate on a much smaller scale. Also there is careful selection, training, and supervision of the field staff. About 3% of the monthly sample is re-interviewed by the supervisors. The interviewer’s reports are checked for incomplete or inconsistent entries. |
|
|
Term
| Is it possible to look at the data and detect bias when it operates evenly across the sample? |
|
Definition
| No. SE’s computed by the half-sample method will not pick up this kind of bias. |
|
|
