Term
|
Definition
| subtract values from mean, square all the differences, add up all the differences, then divide by the number of observations |
|
|
Term
|
Definition
| take sq root of variance. (subtract values from mean, square em, add em up, divide by # observations, take sq root) |
|
|
Term
| steps for measures of assoc w/ 2 ordinal variables (K's gamma, Tau-b, Tau-c, Somer's d) |
|
Definition
| create frequency table w/ cells for totals. id the pairs which are concordant, discordant, tied on the row, tied on the column. For each type of pair (concordant, discordant, TR, TC), multiply the 2 #s together and add up all the products. For ex. (4 x 3) + (4 x 6)... |
|
|
Term
|
Definition
| subtract values from mean, multiply differences for one variable by the the differences for the other variable, add products up, divide by # of observations (not the total number of the two variables added up, but just the number of pairs observations being compared) |
|
|
Term
| how to calc std error for cont variables |
|
Definition
| subtract mean from each of values, square em, add up squares, divide by # of observations minus 1, take sq root, then divide by sq root of the number of observations |
|
|
Term
| two-tailed vs one-tailed hypothesis |
|
Definition
| if population mean could be above or below the test value, it's two-tailed. if the pop mean can only be above or only be below the test value, it's one-tailed |
|
|
Term
| when to use the different means tests -- Z-test and paired T-test |
|
Definition
| use Z-test when trying to find out if a population mean is equal to a particular value. Paired t-test is used to test whether the mean of one variable is equal to the mean of another variable |
|
|
Term
|
Definition
specify the hypothesis that the population mean is/isnt equal to (the test value). Write down the alternative hypothesis and whether it's one-tailed or two-tailed. Calc mean if it's a cont variable. then calc std error. Then plug into formula and take absolute value of answer -- that's the z-score. look up z-score on table and find p-value. if it's a two-tailed hypothesis, multiply p-value by 2.
If p<.05, reject the null with 95% certainty
If p<.01, reject the null with 99% certainty |
|
|
Term
|
Definition
1. Subtract the second variable from the first variable, create a
column of differences
2. Calculate the mean of the column of differences
3. Calculate the standard error of the column of differences
4. Calculate degrees of freedom = (number of non-missing
observations – 1 )
5. Calculate T = ( mean / standard error )
6. Use the T-score table. Find the row that matches up to the
degrees of freedom.
Look up on table. If your T-value is BIGGER than the value for t.025, for example, the means are significantly different at the .025 level for a one-tailed test, and .05 for a two tailed test. If smaller, there's minimal difference between the two means |
|
|
Term
| steps for chi-square test and when to use it |
|
Definition
-use it to test the amt of correlation between two categorical variables -create cross tab. add 'total' column and row. calc expected freqs, plug into chi-square formula for each cell, add up those -- that's the chi-square value. calc degrees of freedom and look up in table.
If your chisquare value is BIGGER than the value for .025, the association is significant at the .025 level |
|
|
Term
|
Definition
-order observations smallest to largest
-Divide the percentile N by 100, and multiply number of observations. If this number has a decimal, round up
- Starting from the smallest value, count up as many observations as the number derived in step 3. The value in that observation is the Nth percentile
***if the number from step three is a whole number before rounding, count up from smallest observation that number and that number plus 1. Add the two and divide by 2 to get percentile |
|
|
Term
|
Definition
-order observations smallest to largest
-multiply # of observations times 0.5
-if decimal round up
-Starting from the smallest value, count up as many observations as the number derived in step 3. The value in that observation is the Nth percentile
***if the number from step three is a whole number before rounding, count up from smallest observation that number and that number plus 1. Add the two and divide by 2 to get percentile |
|
|
Term
| std error for binary variables |
|
Definition
| use 2nd formula under "Margin of Error" on formula sheet (one w/ "p" value) |
|
|
Term
| how to calc margin of error |
|
Definition
-count number who gave each answer (Y) and (N)
-compute the proportion who gave the first answer using this formula:
p= ___Y_____
Y + N
-calc std error using the p-value calc' in previous step
-plug into this formula:
p + or - 1.96 *SE for 95% confidence (replace 1.96 with 2.58 for 99% confidence) |
|
|
Term
| N and m in Kendall's tau-c |
|
Definition
N is the number of observations in the table (not the number of cells). add up all the numbers in the cells to get it.
m is the either the number or rows or columns – whichever is smaller. |
|
|
Term
| How they fool you with stats |
|
Definition
Non-random samples, Improper generalization (Republicans won more seats in the last election, but
that does not necessarily mean voters want more restrictive
abortion law), Small samples / No standard deviations (margins of error)
reported, Self-selected respondents, Loaded questions: |
|
|
Term
| Convert odds ratio to percent |
|
Definition
| subtract one and multiply by 100 |
|
|
