27.10.2014 Views

Russel-Research-Method-in-Anthropology

Russel-Research-Method-in-Anthropology

Russel-Research-Method-in-Anthropology

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Multivariate Analysis 667<br />

In fact, INCOME expla<strong>in</strong>s 43.82% of the variance <strong>in</strong> motor vehicle deaths<br />

(r .662 and r 2 is .4382) and VIOLRATE expla<strong>in</strong>s 6% of the variance <strong>in</strong><br />

motor vehicle deaths (r .245 and r 2 is .0600). Together, though, INCOME<br />

and VIOLRATE have a multiple-R of .762 and an R 2 of .581. Here aga<strong>in</strong>, the<br />

two variables expla<strong>in</strong> more variance work<strong>in</strong>g together than they expla<strong>in</strong> work<strong>in</strong>g<br />

separately.<br />

In other words, it’s the complex association of per capita <strong>in</strong>come and the<br />

level of violence that expla<strong>in</strong>s so much variance <strong>in</strong> both the rate of teenage<br />

births and the rate of motor vehicle deaths. It turns out that lots of th<strong>in</strong>gs are<br />

best expla<strong>in</strong>ed by a series of variables act<strong>in</strong>g together.<br />

Caution: Automated Regression Now Available Everywhere<br />

All of the major programs for statistical analysis today produce what is<br />

called a stepwise multiple regression. You specify a dependent variable and<br />

a series of <strong>in</strong>dependent variables that you suspect play some part <strong>in</strong> determ<strong>in</strong><strong>in</strong>g<br />

the scores of the dependent variable.<br />

The program looks for the <strong>in</strong>dependent variable that correlates best with the<br />

dependent variable and then adds <strong>in</strong> the variables one at a time, account<strong>in</strong>g for<br />

more and more variance, until all the specified variables are analyzed, or until<br />

variables fail to enter because <strong>in</strong>cremental expla<strong>in</strong>ed variance is lower than a<br />

preset value, say, 1%. (There are even programs that test all possible path analysis<br />

equations . . . more on path analysis <strong>in</strong> a bit.)<br />

Stepwise multiple regression is another one of those controversial hot topics<br />

<strong>in</strong> data analysis. Some people feel strongly that it is m<strong>in</strong>dless and keeps you<br />

from mak<strong>in</strong>g your own decisions about what causes what <strong>in</strong> a complex set<br />

of variables. It’s rather like the significance test controversy and the shotgun<br />

controversy I discussed <strong>in</strong> chapter 20.<br />

My take on it is the same: Learn to use all the tools and make your own<br />

decisions about the mean<strong>in</strong>g of your f<strong>in</strong>d<strong>in</strong>gs. It’s your responsibility to do the<br />

data process<strong>in</strong>g and it’s your responsibility to determ<strong>in</strong>e the mean<strong>in</strong>g of your<br />

f<strong>in</strong>d<strong>in</strong>gs. Don’t let robots take over any of your responsibilities. And don’t be<br />

afraid to use all the hot new tools, either.<br />

Examples of Multiple Regression<br />

Graves and Lave (1972) modeled the start<strong>in</strong>g wage of Navajo men who<br />

moved from the reservation <strong>in</strong> the 1960s to Denver. Graves and Lave started<br />

with 19 variables that they thought might relate <strong>in</strong> some way to start<strong>in</strong>g<br />

wage—th<strong>in</strong>gs like age, marital status, father’s occupation, proficiency <strong>in</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!