SPSS PC Version 10: Analysis of Variance (ANOVA)1 ANOVA involves the comparison of means from more than two

groups. The ANOVA procedure results in an F-ratio that tells us the ratio of the variation between groups to the variation within groups. As is true in other hypothesis testing procedures, the purpose is to determine whether or not the differences you observe between the samples represents a real difference between the populations from which the samples were drawn or, alternatively, the observed difference just happened by chance when, in reality, the populations don't differ at all on the variable you are interested in. Using SPSS to conduct an ANOVA test: To conduct an ANOVA test choose Analyze, Compare Means, One-way ANOVA. A dialog box will appear in which you must specify two things: the "dependent list" and the "Factor." The "Factor" is the variable that defines the groups that you want to compare. The "dependent list" will contain all of those variables on which you would like to compare the categories of your "Factors." For example, let's examine whether the average level of education among women varies across the regions of the country. Here our "dependent list" would contain the variable educ ("Education in years") and the "Factor" would be the variable region ("Region where R lives"). The syntax for this is: ONEWAY educ BY region /MISSING ANALYSIS . The results of this test:
ANOVA EDUC Education in Years Sum of Squares 356.025 82954.659 83310.684 df 3 10843 10846 Mean Square 118.675 7.651 F 15.512 Sig. .000

Between Groups Within Groups Total

The output from the ANOVA procedure is actually quite easy to read. In the main box of the output, appropriately labeled "ANOVA," are all of the main pieces of the ANOVA test. Included here are the "sum of squares between groups," the "sum of squares within groups," and the degrees of freedom associated with each. Also included are the "mean squares" between groups and within groups. These are obtained by dividing each "sum of squares" by the appropriate degrees of freedom. Finally, and most important, is the F-ratio and significance level associated with this F(obtained). The F-ratio is calculated by dividing the mean squares between groups by the mean squares within groups. If this value is large enough (larger than the critical value of F) and, therefore, the probability of obtaining an F-ratio of this size or larger (labeled "Sig." In the SPSS output) is small enough (smaller than our chosen alpha), then we can reject the null hypothesis in favor of the idea that the populations we are comparing really do differ from each other. In our example, the F(obtained) is 15.512 which far exceeds the F-critical of 2.60 for this test when using an alpha of .05. Correspondingly, the observed p-value of .000 is well below my chosen alpha of .05. By either standard, I can reject the null hypothesis in support of the conclusion that the level of female education really does vary significantly across the regions of the country. In fact, even if I had used a more stringent alpha level (e.g., .01 or .001) I would have still rejected the null hypothesis. Note that if you choose Options and the Descriptive Statistics in the main ANOVA dialog box, SPSS will also produce an additional box containing the means and standard deviations (and additional information, including Ns, minimums and maximums) for each group being contrasted in the ANOVA test. Although not essential, this is good information to have if you want to look for the source of any significant difference between means detected by the

Prepared by Kyle Crowder of the Sociology Department of Western Washington University, modified by Patty Glynn. November 24, 2000

ANOVA process. It may also be important to look at these to make sure that missing values are coded correctly, and that you do not have unexpected values because of dirt in your data. The syntax is: ONEWAY educ BY region /STATISTICS DESCRIPTIVES /MISSING ANALYSIS . You can also ask SPSS to give you information about what groups vary significantly from others by putting a check in the box SHEFFEE and/or BONFERRONI in the POST HOC box. Bonferroni is a stricter test, which takes into consideration the increased possibility of finding significant relationships because multiple tests are being performed. The syntax and results would look like: ONEWAY educ BY region /STATISTICS DESCRIPTIVES /MISSING ANALYSIS /POSTHOC = SCHEFFE BONFERRONI ALPHA(.05). And the results, as follows, shows that there is a significant difference between those living in the Northeast and South and West, but not between those living in the Northeast and the Midwest.

Multiple Comparisons Dependent Variable: Education in Years Mean Difference (J) Region Where R Lives (I-J) Midwest 4.354E-02 South .4303* West .2882* Northeast -4.3543E-02 South .3867* West .2447* Northeast -.4303* Midwest -.3867* West -.1420 Northeast -.2882* Midwest -.2447* South .1420 Midwest 4.354E-02 South .4303* West .2882* Northeast -4.3543E-02 South .3867* West .2447* Northeast -.4303* Midwest -.3867* West -.1420 Northeast -.2882* Midwest -.2447* South .1420

Scheffe

(I) Region Where R Lives Northeast

Midwest

South

West

Bonferroni

Northeast

Midwest

South

West

Std. Error 8.198E-02 7.622E-02 8.284E-02 8.198E-02 7.064E-02 7.774E-02 7.622E-02 7.064E-02 7.164E-02 8.284E-02 7.774E-02 7.164E-02 8.198E-02 7.622E-02 8.284E-02 8.198E-02 7.064E-02 7.774E-02 7.622E-02 7.064E-02 7.164E-02 8.284E-02 7.774E-02 7.164E-02

Sig. .963 .000 .007 .963 .000 .019 .000 .000 .269 .007 .019 .269 1.000 .000 .003 1.000 .000 .010 .000 .000 .285 .003 .010 .285

95% Confidence Interval Lower Bound Upper Bound -.1857 .2728 .2171 .6434 5.662E-02 .5199 -.2728 .1857 .1892 .5842 2.736E-02 .4620 -.6434 -.2171 -.5842 -.1892 -.3423 5.828E-02 -.5199 -5.6619E-02 -.4620 -2.7356E-02 -5.8277E-02 .3423 -.1728 .2599 .2291 .6314 6.964E-02 .5068 -.2599 .1728 .2003 .5731 3.957E-02 .4498 -.6314 -.2291 -.5731 -.2003 -.3311 4.702E-02 -.5068 -6.9640E-02 -.4498 -3.9574E-02 -4.7017E-02 .3311

*. The mean difference is significant at the .05 level.