Behind the Numbers in Local Races: Updated
This week I was granted access to some of the polling data gathered by a new local communications firm, Out of the Box. One of the men behind the curtain is Richard Korn, former Democratic candidate for Auditor. Richard sat down with me to answer questions about the polling that Out of the Box did for local races, including Insurance Commissioner, SD11 (Deluca v. Townsend), Mayor of Wilmington and County Executive.
If you are a registered Democrat in New Castle County (and you are a likely voter) you probably received a call from Out of the Box. Their polls were probably most memorable for their brevity. Usually just a single question, they got right to the heart of the goal of the poll. “Hello, as a voter in [name the location], your opinion matters. Please answer the following question:” and then the old “if the election were held today…” question. That’s it.
Deluca v Townsend
It was obvious to OOTB where this race was going with a week left. Townsend was pulling over 40% with Deluca hovering around 30% and only 23% still undecided. Certainly the people that commissioned the poll (the poll was sponsored by an organization that supported Deluca) had reached out to Deluca to tell him that things weren’t looking good. But the rumors that I’m hearing from other people about this is that Deluca was in serious denial.
By the time the Saturday before the election rolled around, the deal was done. Townsend had risen in the polls to over 50% and Deluca had dropped to about 27%. The remaining undecideds broke 2:1 for Deluca on election day, but the final tally was to 58-42.
Wilmington Mayor
You wouldn’t know it looking at the comments here on DL, but the mayors race wasn’t even close. Even as far back as mid-August, the race belonged to Dennis Williams. Williams had 28% and a 9 point lead over his next closest competitor (Kevin Kelly). Williams had a base of power in RD1 and 2, whereas Kelly and Montgomery split RDs 3 and 4.
On election day, Williams and Kelly each picked up 40% of the undecided voters. And that is enough to Call Dennis P. Williams “Mr. Mayor”.
In the next post I’ll talk about what the New Castle County Executive and Council President polling showed and when.
Update:
From the comments, I have realized some of the data that I left out. I apologize for that. Here is more detailed information about the polls above.
Deluca/Townsend
Both polls were robocalls made only to likely Democratic voters in SD11 consisting of the same script both times and similar to the one above. The poll consisted of only one question.
September 4 & 5 Total polls completed=175 Deluca: 30.9 Townsend: 42.9 Undecided: 26.2 September 7 & 8 Total Polls completed=175 Deluca: 29.1 Townsend: 48 Undecided: 22.9 Wilmington Mayor This poll was conducted August 11 & 12 via robocall to likely Democratic voters inside of Wilmington. It consisted of a single poll question similar to the phrasing above. Total polls completed=910 Bovell: 8.7 Kelly: 19.2 Montgomery: 17.7 Williams: 24.5 Undecided: 26.9
Interesting stuff. I was wondering how accurate those micro-robo polls were going to be. Apparently, they were pretty accurate. For these races anyway.
Uh, excuse me for casting some doubts on this self-promotion from Richard Korn. No way that undecideds break 2-1 for an incumbent like DeLuca on Election Day, just doesn’t happen.
You might want to ask OOTB what they saw in the Crane race. Korn was involved in that race, and I was assured that the election was all but in the bag for Crane.
Numbers don’t lie. As Managing Partner of Out of the Box, I designed and implemented these polls with Mr. Korn. I understand how difficult it is to make an assessment without access to the data so let me break it out. Trending across 4 polls executed across 4 dates clearly showed Brian Townsend with a commanding lead (between 9 – 15% trending upwards in favor of Townsend). The final poll that was executed on 9/8 (Saturday before primary day) showed 18.2% undecided and a 23.5% spread in favor of Townsend. The election result was a 15.6 point win for Townsend, almost 8% lower than the final poll. The bump in numbers for both candidates (14.9% for DeLuca and 7% for Townsend) on Primary Day would indicate that most undecideds broke toward the incumbent. Townsend, however, had such a commanding lead that this had no effect on the outcome of the race.
Regarding the I.C. contest, no race is “in the bag.” We had data that showed trending toward the incumbent but, because of client confidentiality, couldn’t share that information until now.
Wow, polls delivered after the election. Interesting business model.
Nemski – if you want to know the results of polls BEFORE the election, you have to pay for them. Send Richard a check and we can make that happen.
I agree with El Som here. Big Time. There’s more than a few missing links in Geeks post and interview.
This polling data is worthless without knowing:
A. Who paid for it
B. Sample size
C. Characteristics (LVs v. RVs, party breakdown)
D. When it was conducted
E. How it was conducted (phone, in person, online)
Anyone have the details, or are we just supposed to trust you?
SW
A is somewhat bogus
B varies, but one that I looked quickly at was 3% of the eventual turnout
C Party breakdown doesn’t matter in a primary. They used LV’s with their own formula for those. I am not at liberty to reveal that.
D I did give some dates above “By the time the Saturday before the election rolled around, the deal was done.”
E I thought I was clear about the fact that the polls were conducted by phone.
I will try to better quantify my characterization of the polls in subsequent posts.
ES – Why is it unbelievable that undecideds break for an incumbent? I suspect that is one of the advantages of incumbency.
I have the polling for the Crane race. You will just have to stay tuned for that.
LG:
Sample size is vitally important. Polling 100 is not the same as polling 600. This is a crucial number to gauge a poll’s reliability. Three percent of the turnout in DeLuca/Townsend sounds impressive and substantive, but in reality is just 75 people. No reputable pollster would go with that small a sample. Details matter.
Who paid for a poll is also important. Having that information allows you to adjust for potential biases in reporting of results, timing, etc.
Exactly when a poll was conducted is critical. If calls were done in a day or over a weekend, results will be more accurate. If a pollster spreads them out over a longer period, later responses could be influenced by changing events.
Please don’t let your judgment get swayed because Rich Korn is giving you “special access” to secret polling data.
SW – I understand the need for the polling sample. I will include it in the future. Let me see if I can get pull the sample size/dates on some of these.
I am not privy to who paid for these polls, but the script and the results have been made available to me. The essential script is listed above. There are certainly factors that can adjust the numbers within scripts that CAN influence the poll. I can dig up the polling question and you can discern the bias.
And please don’t ignore things like “on the Saturday before the election” because you have a built-in bias against Richard Korn.
I’m not. But was the entire poll done on that Saturday or did it just finish up then after a week of polling? The difference is crucial.
I have no built-in bias against Richard Korn. He earned that bias by running an inept, incompetent campaign against a guy who should have lost resoundingly. Why we are now supposed to take his insight as gospel is beyond me. So forgive my skepticism about data presented with a gee-whiz, I have numbers!!! approach.
Three percent of the turnout in DeLuca/Townsend sounds impressive and substantive, but in reality is just 75 people. No reputable pollster would go with that small a sample.
It is a small sample for a small race. The PPP Presidential tracing poll surveys a pool of only 1000 Likely Voters, and these guys are about the best in the business. That 1000 is *defnitely* way less than 3% of the likely turnout. Knowing whether they used a Registered Voter vs. Likely Voter model would be useful as would how they determine their sample size.
EDIT: And the MOE — that would be nice to know too.
A small race, yes, but would you trust a 75-person survey?
Frankly, I’m skeptical of most polls until they figure out a way to reach cell-only users.
For a race on the scale of a *Delaware* RD? Sure I would. It would be more important to me to know how they create their sample, to know if they use an RV vs LV model (and how they determine LV) and what the MOE is. A well crafted sample can get you alot of good data.
I’ll get a better representation of these numbers up later today. I have sample sizes, dates, etc. I have unfortunately been trying to look at the data between day job responsibilities. I’ve probably screwed up my numbers within the comments.
Here is the proper numbers on the Deluca poll conducted on Sept 7th and 8th (Friday and Saturday before the Tuesday election). No voters were polled twice:
175 people completed the poll in its entirety.
Question: Hello, as a resident of senate district 11 your opinion matters. If the election for state senator were held today would you vote for Senator Tony Deluca or Bryan Townsend? Press 1 for Tony Deluca, press 2 for Bryan Townsend, press 3 if you don’t know.
TD = 29.2%
BT = 48%
UND = 22.8%
175 voters represents 7% of the eventual turnout on election day.
@El Som: “I was assured that the election was all but in the bag for Crane”.
It was but for McDowell’s two sham candidates. And as far as the pollster claiming it was trending toward the incumbent, I wouldn’t call 32% of the vote a positive trend but a lousy showing that reflects how useless Karen Weldin Stewart is and that people are on to her. So useless, in fact, that the R candidate has a better than average chance unless she does her usual thing, which she probably will, and accuses him of various bullshit irregularities too late for him to do anything about it. But that garbage only works for so long. She’s reached her limit with it.
Someone mentioned KWS. Waiting for JPC, JR. to respond with some hack comment, but maybe he’s all tuckered out from the Connections Talent Show. The wannabe-rappers were hilarious.
Coldwell = JPC Jr? Just askin….
You selling your house if you even own one? Might be a good idea since things are going downhill fast. You didn’t hear any fat ladies sing, did you?
@metoo–No, but the horns group was actually good.
The primary is over ad the city produced some damaged goods for the Democratic party beyond Mayor to be Williams. Gordon has the best shot, his record as Exec was sterling despite his Freebery inspired antics. Bullock with no name recognition beyond a minister from the City won’t be able to get a payday loan after the Delaware GOP gives him a little payback and KWS is well, a bit challenged.
Richard Korn? Really?
I would not interpret the data to indicate that the undecideds “broke” for Delucca. I would interpret it to mean that Delucca’s GOTV effort finally got started, alas too little too late. Delucca got more of his people to the polls than the survey would have predicted. More RV’s than expected.
Regarding likely voters, I think you can purchase voter information from the DOE, which lists which of the last several elections, including primary, special and general you voted in. It’s pretty easy to discern who are likely voters in a primary, if they voted in the last 3 consecutive primaries. If this isn’t available to the general public, it is available to candidates, campaigns and the parties.
It also contains a phone number for the voter, if they provided one to DOE. In my case I registered with my cell, so I get polls on my cell. I can’t be sure, but I think it may also list the voter DOB. I would hazard to guess that a relatively high percentage of primary voters are older and still have land lines.
Maybe we can get together and do some IC polling for the general?
All voter information available to candidates and parties should be available to the general public.
Found it. Phone numbers in voter records are not public. No exceptions in the law that I could find for candidates:
Title 15 § 1305. Custody of registration records; public inspection.
(a) All registration records, when not in the custody of the registrars, shall be and remain in the custody of the department except as otherwise provided. The registration records shall, during normal business hours of each department, be open to the inspection of anyone desiring to examine the same, without fee or reward, provided, however, that Social Security number, date of birth, driver’s license number and telephone number shall not be subject to such inspection. Anyone desiring to do so may be permitted to make copies or partial copies thereof.
I think we should treat this for what it is: Richard Korn trying to tout his latest enterprise. There’s nothing like free media. He has the right to do it. We, meaning I and everyone else, have the right to question the work product.
Maybe he’ll get some clients out of this. Were I running (insert joke here), I wouldn’t be one of them.
This is one of the sillier posts I’ve read on here.
If you hadn’t noticed, there are posts almost every day here where folks are pouring over polling numbers. There is a decent contingent of the DL audience who is actually interested in that — a thing you could tell by seeing the exchanges asking for *more* data. So we’ll look at whatever else comes through on current inquiries and those of us interested in numbers will make out own decisions, I’m thinking.
And Gallup doesn’t do its tracking polls primarily as a public service — it is meant as a business calling card. But while most of us will never engage Gallup for a poll, that doesn’t mean that we aren’t amusing ourselves with the numbers.
El Som, When I run my self-funded vanity campaign against Carper, I’d use them. The accuracy seems to be pretty good and being robo-call based, it must be a fairly inexpensive service.
I think you are letting your dislike of Mr. Korn cloud your business mind.
Yah, playing with numbers is fun. But first you gotta have faith in the numbers and know what they mean. I have zero faith in Korn, and thus begin with very little faith in the data. LG’s breathless stars-in-his-eyes tone – I was Granted Access! – lowers my faith by another notch. And the lack of information about certain details wraps it all up: I’m exceedingly skeptical.
Not to insult LG, but I think he got used big time. Korn is looking to pick up some new clients for the rest of the cycle, so why not reach out with some behind-the-scenes details to a popular blog? Smart marketing. Just not enough questions asked on DL’s part.
If you basically know what you are looking at and have some basic questions answered, you don’t need much “faith in the numbers”. You can figure it out on your own and you don’t need advanced statistics or math to do that. Hence, the fascination with numbers among a certain set here. LG is clearly not done with his series here and a couple of us have asked smart questions that LG is following up on. Sorry this doesn’t interest you, but there’s little need to bore the rest of us with that.
Let me rephrase: I have very little faith in the man behind the numbers. Korn ran a completely incompetent campaign and proved himself to have piss-poor judgment. Why should I trust him to run a competent operation that produces good data?
You obviously haven’t read any of my posts in this thread. I am very interested in the topic, indeed. I simply think you got taken for a ride and that your Granted Access! pal got snookered by a gameplayer. Approach all numbers with skepticism.
And I apologize most profusely if my posts bore you. Should I sing a song to entertain thee?
“Not to insult LG…”
Too late. You were pretty insulting.
It interests me, but the Townsend/DeLuca thing just doesn’t compute.
Here’s why. According to the poll, 23% were undecided in that race, and Townsend had a 10-point lead. In general, undecideds almost always break against the incumbent, and even more ‘always’ with such a disliked incumbent.
In that last week, DeLuca had three, count-em, three negative pieces hit the mailbox on the same day. It was a mistake, but one that was not going to benefit DeLuca. In the same period, Townsend mailed his only aggressive contrast-and-compare piece and a personal letter.
Plus, there was no way that DeLuca’s GOTV was better than Townsend’s. Meaning, as far as I can figure, the initial polling was much too favorable on the Townsend side.
Granted, there could have been a ‘Rizzo’ effect, named after Frank Rizzo, who people voted for but didn’t want to admit that they voted for. Good polling is designed to account for that effect. Which this poll was not.
And they can mealy-mushmouth any work that they did for Crane all they want, but I KNOW what that polling showed, and it was wrong. Which, for me, is one more reason to look askance at anything that Richard Korn does.
Oh, good. He deserved some insulting for putting out such a detail-free post.
By the way: A sample size of 175 people will generally result in an MOE of between 7 and 10 percent. (http://www.publicagenda.org/pages/best-estimates-guide-sample-size-and-margin-error). That would render the vast majority of local race poll results meaningless.
I have very little faith in the man behind the numbers.
So what? If you know some basic information about the numbers and how they were gotten, you don’t need faith. You just need the math.
So why do you folks here moan and groan over polls conducted by allegedly left or right leaning pollsters? Who conducts the poll does indeed matter. I’m sorry that you’re so defensive to recognize that fact.
Who conducts the poll does matter, but you can point to real problems with data/methodology/disclosure in order to make your argument. Not just that We Don’t Like Them.
So sorry that these polling conversations have been going over your head.
You’re right. I should conduct thoughtful analysis like Delaware Dem:
So I have swung back to Jason330’s line of thinking: frack Rasmussen and other Republican polls whose sole purpose for existing is to shape narrative rather than conduct scientific polls. They cannot ever be trusted. And if we are going to criticize people like Allen Loudell for trusting Rasmussen, then we cannot very well use their polls in our polling report. So Rasmussen and We Ask America are gone.
…
Yes, this a Rasmussen poll, and yes, I am going against my own rule and counting it in the map. Why? Well, as we all know, Rasmussen has a Republican bias. So if their biased poll is only showing a 3 point lead for Romney in Missouri, then I think we should pay attention to it.
One of the real problems that pollsters have is in simply reaching voters now that we have cellphones and call-screening stuff.
I sometimes respond to polls and I sometimes don’t. Meaning that I’m an unreliable respondent. I usually tell the truth, unless I’m trying to figure out who is behind the poll, in which case I might try to get the ‘push’ questions, also making me an unreliable respondent.
To my knowledge, this Korn outfit was not polling for either side in the DeLuca/Townsend race, so they’re pointing to some alleged results in a race that they were not polling to make some statement about their credibility. With or without polling, I think I had at least as credible a take on this race. Well, actually a MORE credible take on this race.
Again, I have no problem with this outfit trying to get some free media attention. But it’s not gonna come with a suspension of disbelief on my part and on the part of other commenters.
There is no one here asking for a suspension of belief. If you have enough information to write this off, that’s just fine. I do not have enough information — because I have questions outstanding AND LG isn’t finished here. So what does it cost you to let the story play out? Oh, I forgot, one more chance to tell everyone how little regard you have for Korn. And frankly, I don’t much care about that, but there are people who are trying to take a look at whatever might be here to give it a fair assessment.
On the DeLuca/Townsend GOTV thing — I know from someone at ground zero of his GOTV effort that they were furiously dragging people out to vote all day. And they had a fairly remarkable cohort of folks helping him do it. It was too little too late, (and they seemed to know it late afternoon, early evening) but they sure did have a big effort at the end.
Just my little opinion about undecided voters. I knew I was going to vote for Townsend. But if I was undecided, the 8 or 9 Townsend yard signs in my rather small neighborhood probably would have swayed me.. because generally I like my neighborhood and trust my neighbors. Yard signs mean something to me, the signs out in intersections, not so much. (there was one Deluca sign in here)
OK, Cassandra, I’ll bite. Everything I was told (and quite possibly that you were told) was that Crane’s people had him in front by a reasonable margin. This was about 10 days before the election.
Now we apparently find out that he was using a first-time polling firm headed up by Richard Korn to do the polling. That’s political malpractice.
And maybe THAT’s why he never bothered with even a half-assed GOTV in New Castle County. The polling was fucked up.
Might I humbly suggest that people gaming the poll in search of a sponsor could have contributed to the result you disbelieve in the DeLuca-Townsend race?
Might I further humbly suggest that compared with some of the other Delaware “polling experts” out there, this poll has by no means proved useless.
Even if this is a new business venture backed by someone in politics who rubs a lot of people the wrong way, it is possible to judge it on its merits. Do we hold the design of people’s yard signs against Charlie Copeland?
“there are people who are trying to take a look at whatever might be here to give it a fair assessment.”
And no one is doing or saying anything here that would prevent or hinder them from taking said look. Why are you so worried about them?
Not bothering with a GOTV in NCC isn’t the fault of polling.
SW: Let’s see how much they release on the particulars. If they’re trying to be a real polling firm, they’ll be fairly transparent. Holding your breath?
True, Pandora. But a pollster lighting a fire under the campaign’s butt by providing them with accurate numbers might have, emphasis on MIGHT, made a difference. Polling is used by campaigns to determine strategy, not to provide them with some predicted favorable result. Hard to imagine any pollster not identifying the Crane lag in NCC in their polling.
I just don’t get not bothering with GOTV. Obama is up in many states, should he not bother campaigning, organizing or getting out the vote? This strikes me as the candidate’s misstep – an over-confidence thing that we all seem wary of on the National level.
Update below from Jim Martin on David Brady’s Voter’s Guide for the 1st District that turned out to be ‘falsely labeled’ and was actually voter data for the 3rd District. Waiting to hear back as I hope to get to the bottom of this soon for good or ill:
“Here is what my computer expert said about whether Brady’s list was mislabelled deliberately or simply in error:
There is a field on the Excel data which is for Senate District. It’s called “SD”. I didn’t bother to print it out on the pdf file I sent you as it was always “01” (see attached image where I’ve highlighted this field in yellow). I wonder if his data file shows a “01” in that field or if it shows “03”. If his data file shows “03”, then the mistake made was one of mislabeling. If the data was also changed from “03” to “01”, then something untoward was afoot.”
I was told that polling had KWS safely ahead for weeks prior to the election. Someone out there correctly predicted the KWS, Bullock and Gordon’s margins.
Pandora: The one thing that you can count on w/Obama is that his microtargeting/GOTV efforts that were so successful in 2008 will be even more successful this year.
For example, and I think this is in Iowa, the Obama campaign has requested something like 100,000 absentee ballots and the Romney campaign has requested something like 16,000.
And they’re really working their field magic in ‘early voting’ states. The Obama campaign has always seen this level of GOTV as their secret weapon, and have especially targeted it in light of the anticipated Romney advantage in TV advertising, a perceived advantage that has not yet manifested itself. So, right now, the Obama campaign is harvesting early votes in states at a time when polls show Obama with a pretty decent advantage.
Come to think of it, I’m starting to suffer from overconfidence.
Even if this is a new business venture backed by someone in politics who rubs a lot of people the wrong way, it is possible to judge it on its merits.
Geezer is right across the board here. If they can’t show us their homework, that isn’t the kind of marketing they are looking for, I’m certain. And their work can’t pass muster here OR the transparency is lacking, that is marketing that exists on the Google for a really long time.
And, of course, there is no possibility that a poll in your favor doesn’t tighten up in 10 days. Sheesh, it does happen. I don’t know about the GOTV effort for Crane — it is possible that they were working that with the coordinated campaign. But like I said, I don’t know. One thing I DO know — there was a good six figures of street money with a locus in the 1st and 2nd in Wilmington. No matter who was doing GOTV that was going to be a tough thing to overcome.
Every polling company starts someplace. I don’t know who is doing the polling (it isn’t Richard Korn) or that person’s background. But the art in polling is mostly around constructing the samples, constructing and testing your voter model, constructing your questions. Much of the rest is fairly straightforward math.
If his data file shows “03″, then the mistake made was one of mislabeling.
Or it was a mistake of not looking at your file. It seems to me that you would at least look for your own address. But most folks are winnowing their voter lists to get a better bang for their buck (does anyone send a mailing to EVERYONE on their voter lists?), and I can’t imagine that you wouldn’t notice at least that there are zip codes missing or buildings or streets that shouldn’t be on your list.
Totally off topic:
Whenever I read a post by cassandra_m, I read it in the voice of Michel Martin.
Think James Earl Jones. With the same intimidation factor…
SW – “By the way: A sample size of 175 people will generally result in an MOE of between 7 and 10 percent. (http://www.publicagenda.org/pages/best-estimates-guide-sample-size-and-margin-error). That would render the vast majority of local race poll results meaningless.”
THANK YOU THANK YOU THANK YOU!
Jason330 – I hope you recognize how shoddy their pollsmanship is as more details emerge.
El Som – You might want to see if Mitch asked for his money back.
Well, PM, i hate to tell you that your link is irrelevant. It applies to a universe of 209 million adults. A sample of 175 from a universe of a few thousand will be pretty accurate if repeated, properly weighted, and duplicated. If 4 thousand people are voting, you should not expect to need 1000 people in your survey, a couple of hundred would do.
E. S. it isn’t about the person. LG hasn’t given us a lot of information. I think he showed to be pretty accurate. We are not expecting margin of error to 2% in a local race. We want to know if we are winning, if our ads are effective, if the opponent’s attacks are effective, and if our people plan to vote. It is a myth that late undecideds go to the challenger, they tend to split or stay home. Sometimes they go to the best closing case. It is true that earlier undecideds tend to go to the challenger. If this poll is to be believed, that happened. When he trended over 50%, the rest was academic. Delucca’s last minute push helped, but it could not beat momentum. Btw 80% of the time a majority of the overall undecided vote goes to the challenger.
” I don’t know who is doing the polling (it isn’t Richard Korn) or that person’s background.”
Good gravy. Why do we bother to listen to people who can’t Google? http://ootbusa.com/?page=about
Awesome. Are you looking at some secret information at that Google page? Because I certainly don’t see any detailed background data on anyone with a statistics or polling background.
The file markings are not easy to spot.
Doubtful scenario: A busy-as-hell candidate deciding to spend the time to search for their own address — to double check Voter Guide handed to them by State DoE staffers — ?
Dave Brady may not even have the computer capacity to search through a huge file like Voter Guide even if he had had the unlikely suspicion, and in this case, foresight that the State of Delaware had to be second-guessed.
I know I don’t have the capacity to search through a file like that.
I just wanted to drop in here to say that my day job has had an unexpected surge in its need for my attention. I haven’t been avoiding the promised updates, I’ve been sleeping and working (not necessarily in that order).
That’s exactly why I under promise and under deliver.
OK, post updated, do your worst.
MOE?