25May/10Off
Anova or Kruskal-Wallis? 10 points?
I am unsure of which test to use...
I need to formulate a null and research hypotheses as to whether rates of Y (auto deaths per 100mill. miles traveled) differ by region of the country.
In SPSS...the REGION variable has 3 "values" (northeast, southeast, west) and is a "scale" measurement.
The Y variable (auto deaths) is also a "scale" measurement.
the sample i am using for Y is "Traffic Fatalities by State" With a total sample size of 42,636 (from each state in the USA)
http://www.census.gov/prod/2006pubs/07statab/trans.pdf
May 25th, 2010 - 10:54
REGION cannot be scale. You have to change it to Nominal.
(Correction)
I just realize the fatality rate has already been provided. You only need Table 1083: Traffic Fatalities by State.
Your homework is still to assign which REGION a state should belong to. You have a total of 51 cases. You should run exploratory analysis to see if your three groups (regions) have normal distributions. However, since your smallest case still have a sample size of 43, I would simply assume it is ok to run ANOVA.
May 25th, 2010 - 11:23
Test for the normality of the data. If the test shows the data are normal, use ANOVA; otherwise, use Kruskal-Wallis test.
Since you have more than 40,000 values (for each region?), If SPSS has a procedure to test for normality, you may use it. You can also draw a frequency histogram of each set and see if it resembles a bell-shaped curve.