Eli Boettger | @boettger_eli | 12/11/2017
Sometimes it’s just best to tell people that they screwed up.
That’s the lifeblood of “Poll Attacks” — a weekly column by CBS Sports college basketball writer Gary Parrish — which lays into AP Top 25 voters every Monday for providing ridiculous ballots. Parrish introduces the pollster with the outlandish ballot, summarizes what he/she did wrong, and explains its absurdity.
Harsh? Barely. Necessary? Without a doubt.
First off, Parrish’s column is the only true checks and balances system behind the AP poll. And from what has been dug up by Parrish in the past, some individual rankings have been borderline troubling. This includes AP voters moving up teams after dropping multiple games to unranked opponents, placing teams ahead of opponents that they blew out the previous week, and even accidentally inputting Mount St. Mary’s instead of Saint Mary’s. Everyone has a bad week, but Parrish has discovered a bad poll every Monday for more than five years now.
I decided to parse through over five Google pages of Poll Attacks from the past, gathering well over 30 polls from writers who landed in Parrish’s weekly columns. One thing was made very clear: pollsters pay way more attention to their rankings the week after being attacked.
This shouldn’t come as a surprise. Parrish has an obviously large following — well over 150 thousand Twitter followers — so a sizable chunk of the college basketball universe knows when you messed up. Maybe more importantly, though, is that no credentialed journalist wants to look like an idiot. Well, almost.
Calculating the strength of an individual poll is relatively simple. I pulled up the writer’s poll, lined it up next to the official AP poll, found the absolute value in difference of each of the writer’s 25 rankings, and then found the average of all 25 picks to obtain the poll’s average error.
On average, those who landed in a “Poll Attacks” column had an average inaccuracy of 2.7 rankings in the poll that was being blasted. This means that each ranking was off by about 2.7 rankings, on average, compared to the AP poll that week. The following week, the average improved to 2.33 (a smaller error is better). This concludes that, on average, after being mentioned in a “Poll Attacks” column, writers were more accurate with their next week’s AP poll by 0.37 rankings per selection.
Parrish’s weekly attacks are clearly working. Of those who have been mentioned in a “Poll Attacks” column in the past five years, a whopping 77.8% of pollsters had a lower average error in the following week’s poll. It’s impossible to know how many of the attacked pollsters read or were influenced by Parrish’s columns, but more than three-quarters of pollsters had a stronger top 25 the week after their infamy.
Let’s take a look at Los Angeles Daily News’ Thuc Nhi Nguyen’s polls from earlier this season.
She appeared in the “Poll Attacks” in the second week of the season after ranking Texas A&M six spots behind West Virginia, a team the Aggies had dismantled days prior. She had A&M 23rd on her ballot that Monday and West Virginia 17th, while the AP ballot had the teams 16th and 24th, respectively.
Nguyen responded nicely, though. The following week she had Texas A&M 16th in her poll and West Virginia 22nd, and the two teams were 16th and 23rd in the AP poll, respectively.
The graph below shows the error by pick for each of Nguyen’s 25 teams for Week 2 (blue) and Week 3 (gray). In Week 2, the poll that was under attack by Parrish had five teams ranked either five spots higher or lower than the AP poll. In Week 3, Nguyen didn’t have any teams that were either five spots higher or lower than the AP poll, and just two teams that were three spots higher or lower than the AP poll.
Not everyone seems to catch on, though. In Week 3, Carlos Silva of the Lubbock Avalanche-Journal was attacked for slotting Northwestern 16th after losing to Creighton at home, sneaking past La Salle, and then getting throttled by Texas Tech by 36 points (a team he had ranked 23rd the same week). His Week 3 poll had an error of 2.68 but the following week was even worse, notching an alarming 3.16 error, a net difference of -0.48. It doesn’t happen often, but there have been a handful of pollsters who have failed to bounce back after landing in “Poll Attacks.”
So what does this all mean? First, there is a major voice in the sport openly criticizing some of college basketball’s most respected writers for inexplicable ballots. Secondly, and most importantly, is that it is clearly making an impact.
There really isn’t any reason why “Poll Attacks” should exist. One would think that every AP voter ensures his/her poll is, at the very least, comprehensible on a weekly basis.
Most do a great job.
Others get blasted.
Now head over to CBSsports.com to see who wasn’t paying attention this week.