http://images.latimes.com/media/acrobat/2003-08/9130937.pdf HOW THE POLL WAS CONDUCTED
The Times Poll contacted 1,351 California registered voters,
including 801 voters deemed likely to vote, by telephone Aug.
1621. The margin of sampling error for both registered and
likely voters is plus or minus 3 percentage points. To allow for
analysis, the main sample was supplemented to provide a total of
125 Latino likely voters (margin of sampling error +/ 9 percent-age
points) and the samples were weighted to their state propor-tions.
"...(margin of sampling error +/ 9 percent-age points)..." That means if they polled Arnold at 22% then his support actually could be ANYWHERE from 13% up to 31%. If Cruz polled at 35% then his support actually could be ANYWHERE from 26% to 44%. Since they overlap, the poll is meaningless as they both are within the margin of error of each other. Arnold could be at 31% and Cruz at 26% given the +/- 9% margin of error. Most polls cite a +/- 3% margin of error. That this one is so big means they knew they were skewing the results.