Regression and table hints

1- When it is a table of percentages, nicer to display the percentables as decimals, and indicate in the title or heading that is percent, rather than reproduce % sign for each number. Remember that 43% = .43

2- Be careful with "affect" and "effect". In economics "effect" is used as a noun ("The effect was large." and "affect" is a verb ("The explanatory variable did not affect the outcome.")

3- Whne using dummy variables and interactions with dummy variables, you have to include in the regression both the dummy alone and the interaction; you can;t just include the interaction term and omit the dummy. i.e. time = oxen + bwa + bwa*oxen, and not time = oxen + oxen*bwa.

4- Generally in tables you do not need column or row totals. They clutter and distract!

5- Regression results that are not significant are not "unfortunate"... they just are what they are. (So don't write a sentence like, "Unfortunately, the coefficienct was not statistically significant.")

6- When discussing regression results always use the same, or very similar, formula: "The estimated coefficient was 2.75, which means that a one unit change in xxx would cause a 2.75 change in yyy, and it was statistically significant. This effect (not affect!) is quite large/small."

7- In most economics studies, an R-square of .14 is perfectly reasonable, for example. There is a lot of idiosyncratic variation across individuals and households and nations. There is no "cutoff" R-square.

8- The intercept in a regression is not the average! Draw a scatterplot of the data as we have done in class, then draw a regression line- you will easily see that the intercept cannot be the average of anything.

9- When you are doing crosstabs, it is good practice to do some t-tests in Excel. it will get you in the habit of always checking to see if two means are statistically different. Remember you can only do the t-test when comparing two groups. For other types of comparisons you use ANOVA or more complex tests that you can learn from stats courses.

10- In a regression analysis you cannot include categorical data (generally) as a variable. You have to convert to dummy variables. I.e. Ethnicity= 1 for Mossi, 2 for Bwa, 3 for Djula. Including ethnicity is meaningless- what does it mean to increase your ethnicity by 1 unit? So you have to convert to dummies- 3 dummies, of which you only include 2 (you always exclude one dummy category.)