Datanálisis throws in the towel (UPDATED)

Want answers? Search elsewhere

Reputable Venezuelan pollster Datanálisis has just leaked its final poll before the election. The results are jaw-dropping, but for the wrong reasons.

Their poll, leaked to Bloomberg’s Charlie Devereux and Dan Cancel, has Chávez at 47%, Henrique Capriles at 37%, and get this … roughly 15% of voters are still undecided.

So, let’s walk through the scenarios…

Suppose on October 7th the 15% undecided abstain. Chávez gets 56% and Capriles gets 44%. That is perfectly consistent with the poll! Datanálisis comes out looking great.

Suppose instead that Chávez gets 52% and Capriles gets 48%. Datanálisis can then come out and say most undecideds broke for Capriles, but not enough to stem the tide. Their reputation is intact.

Suppose instead that Capriles gets 52% and Chávez 48%. Datanálisis then says that most of the undecideds went with Capriles, as they thought they would. Veinte puntos!

In fact, there isn’t a single plausible scenario where Datanálisis is proven wrong. Every possibility is consistent with their poll.

I dunno, but the whole point of a poll is to shed some light on what’s going to happen. There must be some sort of prediction we should be able to get from these numbers.

What am I missing here? Are these all voters or likely voters? If it’s all voters, then why aren’t we looking at numbers for likely voters? And if it’s likely voters, how can there still be 15% undecided? I guess there is some technical issue here that I’m missing. It would be great if we had more information…

In the meantime, I am just not buying this at all, at least not at this stage of the game. This isn’t a poll, it’s a tautology.

UPDATE: A couple of readers whom I trust chime in. They say that Datanálisis is actually predicting a Chávez win, and that the large number of undecideds show up mostly when you look at all voters and not likely voters. Apparently, once you narrow it down to likely voters, Chávez’s advantage remains at 11 percentage points. They say that a Capriles win would clearly make Datanálisis’ clients believe the firm made a mistake. I’ll post an update during the day, as apparently several pollsters are giving out final numbers.

86 thoughts on “Datanálisis throws in the towel (UPDATED)

  1. I’m perfectly willing to eat my words on this one … digan que gana uno o gana el otro, pero dejen el guabineo!

  2. Venezuelan President Hugo Chavez is the 4th most popular president in the Americas, according to a new study of presidential approval ratings in the region.

    The study, by Mexican polling firm Consulta Mitofsky, gives President Chavez a “high” approval rating of 64%, gaining 6 percentage points since the firm’s last study and jumping up the table of presidential popularity levels.

    The findings come less than two weeks before Chavez seeks re-election on October 7 against right-wing opponent Henrique Capriles Radonski.

    Obama is 10th…

  3. 15% undecided? That’s damn high anywhere, even more so in a polarized climate like Venezuela. Either tons of people don’t like answering polling questions or there’s something quite amiss.

    • I was talking with a friend today and she mentioned that in this highly polarized climate, another option would be that people would answer that they were likely to vote because they were ashamed (or a similar emotion) of confessing to not voting. If that is so, it might be plausible that they are simply abstainers unwilling to say it.

      Of course it is also as possible that they don’t want to tell the pollster their candidate.

  4. I strongly disagree that polls ought to be able to be predictive. That’s like saying my car ought to be able to fly over traffic jams: it would be awful nice, it would fulfill a deeply held emotional need of mine…but it’s just not what the technology is able to do.

    Polls are by nature backward looking. Not this poll or that poll, polls as such. We’d all love to endow our pollsters with clairvoyance because we find the emotional toll of uncertainty so hard to bear. But that pollsters aren’t clairvoyant is our problem, not their problem.

    • Sure, polls are not supposed to be predictive. That’s why people pay millions of dollars for them – to know what a small sample of the population may or may not have thought a few weeks ago about an important topic. They’re a history lesson! Don’t go making any inferences from them!

      • LVL goes out and asks 100 people how they intend to vote.

        47 say they’re going to vote for Chavez, 37 for Capriles, and the others say they haven’t made up their mind.

        LVL turns around and says:

        “Hey everyone, 47 people told me that they want to vote for Chavez, 37 said Capriles, and the rest said they haven’t decided…”

        …and somehow that makes him chickenshit for refusing to take a stand?!

        I just think that’s weak.

        The data is the data. There’s no use getting mad at it.

        • Ha! But there is useless data too. Crunching numbers is the easy part of doing polls. Now how “LVL goes out and asks 100 people how they intend to vote” makes a ton of difference. Having a 15% of undecided tells me that he is not doing it right.

        • I think that most people are being way to hard on Leon. He could just assume that the undecided will abstain or maintain the same tendency of the decided and predict a Chavez win with no warning. If Capriles wins, it would be a good defense for him to argue that it was impossible to know for whom they undecided were going to vote for (like most pollsters in Nicaragua when Chamorro won) because its true, there is no honest way of knowing what the undecided would do, his poll is not the only one getting a high percentage of undecided voters. I think he is not taking the easy way, maybe he is even trying to throw a bone at the Capriles camp.

        • I didn’t say Datanalisis didn’t get what it’s releasing to the public. My problem is with your claim that polls are not predictive, which is exactly what LVL keeps saying. That’s BS to me.

        • What polls should be is an accurate snapshot of a moment in time. And it is not accurate to state that at this moment in time 15% of the Venezuelan electorate is undecided. Period.

          • With all due respect, this is a lot of baloney.

            Guys, maybe it’s just hard to measure accurately where Venezuelans stand! I actually think it’s more dishonest to speculate about the preferences of ninis. Anybody who’s taken a lab class know that underestimating the error in a measurement will earn you a smack on the wrist.

            • Hence, the title of the post. Datanálisis seems to have given up on trying to decipher what’s gonna happen! Ahora todos somos indecisos…

        • It is a fact that there is an unusually large number of undecided voters in this election, according to Datanálisis. More so than in comparable situations in other countries. That could mean one of two things:
          a) there are actually more people who are undecided now, more than in, say, France or the US.
          b) Dataálisis didn’t ask the questions correctly, didn’t sample correctly, or didn’t follow up enough to infer where these people are going to vote in a few days.

          Take your pick as to which answer is more satisfying.

    • Do you know what a persistent stochastic process is? I guess you know since you even write about Laws of Physics and Economics.

  5. (1978 elections, Luis Herrera vs. Luis Piñerúa)
    Un señor de Margarita
    llamado Ñero Baruta
    se presentó donde el brujo
    para hacerle una consulta
    ¡Perdone usted la molestia!
    Que a preguntarle yo vengo
    que debo hacer señor brujo
    con el problema que tengo
    soy padre de cinco hijos
    y muchas obligaciones
    pero señor no me olvido
    que ahí vienen las elecciones
    dígame usted señor brujo
    por quien yo debo votar
    pa’ cumplir con mis deberes
    de ciudadano ejemplar
    unos dicen que es de Oriente
    otros dicen que es Central
    y yo me encuentro metido
    en un gran berenjenal
    ¡Dígame usted por favor!
    El nombre y el apellido
    si son gordos, si son flacos
    o si son bien parecidos
    Para empezar la consulta
    corta una baraja asi
    Por ti, por tu casa
    por lo que espera
    ¡Vuela Satanas!
    Yo digo lo que veo
    Hay mijito lo que pides
    no es fácil de contestar
    pero mi bola de vidrio
    algo te puede explicar
    Vamos a ver
    Aqui veo un nombre
    el nombre es Luis
    Si señor
    Luis se llama el candidato
    que va a ser el ganador
    el apellido mijito
    en mi bola no se ve bien
    pero tiene siete letras
    siete letras puedo ver
    si son gordos o son flaquitos
    nada de eso veo yo
    lo que si se ve clarito
    es que son feos los dos
    lo que si se ve clarito
    es que son feos los dos

  6. Since being a scam artist seems to be profitable, I think I am going to give it a shot. I have spent the last twelve years studying them.

    So, I give a shot, you let me know how am I doing:

    Va a a ganar un candidato…
    apoyado por el pueblo, que tiene un sentido del deber profundo.
    Sus enemigos lo odian a muerte, pero él es la esperanza del país
    Sus iniciales son H C

  7. A Capriles’ victory is practically impossible according to this poll. In order for them to be TIED 83 percent of undecided voters should go for capriles. As for who these “undecided” are, LVL has said they might even be abstentionists, which makes it even less likely that capriles wins. Again, according to this poll

  8. As Christiansen (Consultores 21) put it recently in a forum on AS/COA, any prediction that projects to a vote of less than 6.5 million (of an aproximate total of 14 million) for Capriles is highly suspicious. In Event #1 (undecided split evenly) Capriles would get 6,1million. In Event #2 (All undecided turn against Capriles) he would get 5,1 million. And in Event #3 (all undecided vote for Capriles) he would get about 7,2 million….well, guess who predicted Capriles would get around 7million votes? Aristóbulo Istúriz himself.

  9. In the recent Québec election, we had a similar situation. It turned out that the “undecided” were liberal voters. When people said they were undecided it was because they were going to vote for the “very unpopular” liberals and did not want to tell. It is called “la prime à l’urne” of the liberal party.

    If I extrapolate that to the Venezuelan situation, I tilt towards labeling the “undecided” as chavistas: if after 14 years of autocracy, corruption and inefficiency you still label yourself as undecided, it is because you do not want to say you’ll vote for Chávez anyways.

    I hope it is not the case, but unfortunately that is what I think.

    • This is a very interesting point.

      Being here in Caracas I see that as difficult to believe but not entirely impossible.

      Unlike Quebec, the dependence on government is higher and the presence of the government in your private life is also higher (or maybe not). I can hardly think of somebody being embarrassed of speaking out loud about voting for Chavez.

      • Probably some people belonging to the middle and upper clases who vote chavista (even in the most sifrino centers they get 2 or 3 percent of the vote) might be afraid of identifying themselves as chavistas or as voting for Chávez. But I think that the statistical impact of this “fear”, in case it exists ,its very low or non existent, while people lying about voting for Chavez or saying undecided instead of chosing Capriles might have more of an impact.

      • I haven’t been in Caracas in a long time so I won’t argue. Anyway, we, Venezuelans, are highly emotional in our decision making. If Capriles manages to create enough momentum in the week ahead he might win… who knows. But make no mistakes, the still undecided will follow their hearts not their minds.

  10. only datos is worthy of even looking at it’s numbers… so LVL can just take a walk on the dark undecided side

  11. I agree 100% with JC on this one. This sounds awfully fishy on the part of Datanalisis. They are clearly hedging their bets with this undecided business. I understand this is like the third poll done by Datanalisis about this election where they get such a large number of undecided people. Would it not be logical that if they really got such a high proportion of undecided voters the first time they would try to adjust their sampling or questionnaire to try to get some idea about the likely preference of those undecided voters? That is the sound and professional thing to do. Instead they choose to put out this Delphic poll that is completely useless. If you poll 3000 people and 450 of them don’t commit to any relevant option then you change the phrasing of the questions, the order of the questions, you train pollsters differently, you do something! You don’t repeat the same useless strategy unless you have a hidden agenda. To me Datanalisis lost any credibility they had.

    • I think I agree with your analysis but disagree with your conclusion. The liberal vote was hidden because people were EMBARRASSED to admit they’d vote for Charest. But the government in Venezuela has spent millions of dollars stigmatizing support for Capriles in Class D and E circles. To the extent people in Venezuela would be embarrassed to admit who they’re voting for amid their peers, I think it cuts the other way.

      But, again, I plead agnosticism. I don’t think we can know who’s right before 7O.

      • People in Venezuela who won’t/don’t state their opinion are doing so almost entirely for fear of Chavista reprisal. And, for the same reason, as Salvador Seekatz says above, there are those who will say they’ll vote for Chavez (e. g., Government employees), even though they can’t stand him and will vote for Capriles.

  12. DIscúlpenme, pero bien pendejos que son quienes le siguen cryendo a cara de queso. ¿Acaso no se dan cuenta de que este tipejo está ocultando el voto caprilista en sus indecisos?

  13. Descifrado has an interesting report on the “indecisos” topic on their weekly letter, according to a source consulted on the issue (Yván Serra), apparently using data from previous datanalisis polls during this election, undecided voters could be estimated to be 83% in favor of Capriles, i’m not one to debunk this theory, but it could be worth analyzing..anyone else read this week’s Descifrado letter?

  14. If you ask the wrong the wrong question, you get the wrong answer.
    Consultores 21 asks: Who will you vote for on October 7th?

    Hugo Chavez
    Henrique Capriles

    Datanalisis asks (I hope I remember all the choices):

    Hugo Chavez
    Henrique Capriles
    Otro
    Ninguno
    No se

    The reason is historical, that is how they have asked the question in all elections. As you can see, it leaves a lot of room for error.

  15. Miguel is right. Besides, after you have answered a a rosary of questions, you have to give your name, age, phone number, and suburb where you live. Datanalisis uses this info to check if you are really you. They pick some questionnaires at random and actually call and confirm the data. The questions of personal data come at the end and that’s when people prefer to remain silent or simply won’t tell the truth. It is frustrating for the interviewer because if the data cannot be checked he won’t get paid. (I think they pay Bs.50 per survey, but it depends on the number of questions). Please correct me if I’m wrong.

    • Yor’re actually describing the supervision say method of all survey firms.
      Folks, keep the eye in the ball. We are at two weeks from the election and Chavez
      is not winning. This is unheard off. He needs 50% plus 1 to win. And he is not there. Plus, his trend in all polls is downward. See the poll from the historically pro government IVAD released yesterday. They have Chavez right at 50% after consistently having him way above.
      In the lack of additional data I can omly speculate that the undecided are the proverbial Ni-Nis. Ni-nis basically don’t love Chavez but saw no alternative in the opposition in the past. But this time I think Capriles got them thinking. So, the game ain’t over till is over.

      • I’m not great at math, but to win Chavez needs 50% plus one of actual votes cast in the election, not 50% plus one of voting intention among those who say they are certain to vote. It’s not the same thing.

  16. I don’t think the amount of undecided voters is wrong. I think there are a lot of people, who voted for Chavez before, or working in the public sector, who are still afraid of voting against Chavez, and that are still not convinced on the secrecy of the vote or feel somhow indebted to him for whatever benefit they have received from his programmes. People who sort of like Capriles’ offer and approach better, but used to vote PSUV, and their families and friends vote PSUV. I think Oppo campaign has been very successful in inspiring doubt in these voters, and they may, on the day itself, follow their gut and surmount fear. I think nobody will know how it will turn until the results come in.

  17. These guys bring out the same poll every election except that in the last AN election opposition won the popular vote. Same with the referendum to change the constitution to allow Chavez to do what he wanted that he did anyway. His numbers have not gone up they have gone down. These polls are as accurate as taking a poll by counting the number of Chavez Bumper Stickers (or sign)vs the Capriles Stickers on a given street corner in a number of Venezuelan Cities. I think the polls are to try to mirror the CNE’s algorithms whenever they get a chance to run them.

  18. Breaking up is hard to do… If you had sided in the past with Chavez, it may not be so easy to take the step to go the other way. I don’t think it is only fear, I think being “undecided” is a transitional label that allows them to test the waters and leave their options open waiting to be convinced.

  19. Leaving aside the question of the undecided for a second (but not more than that), interesting to note is the jump in capriles over the last month of around 5%. Specially considering that for the 4th or 5th time straight chavez has polled somehwere between a 46 and 48. Chavez hasnt been gaining new voters for a while, capriles instead is able to increase his support a month into the elections.

    Going back to the undecided, it is not the first time datanalisis polls show such a high number of them so close before an election. In fact in 2006 the last omnibus before the vote (just a couple of weeks before) put the number of indecisos at around 15%. A posteriori, survey analysis indicated that actually none of the indecisos voted. Are 2012 indecisos in a comparable situation? Methinks not. I have also confitmed with datanalisis folk hat roughly 80 of them have shown an inclination for capriles in the last few months. To what extenet will the indecisos actually mobilize in accordance to these assigned preferences? That is the question. And i share quicos frustration that it is such an open ended one.

    Lastly I recommend people check out smilde’s article. I know he has been criticized in this blog before, but maybe this post can show why such criticisms Might have been a bit too harsh last time.

    http://venezuelablog.tumblr.com/post/32179796880/consultores-21-projections-of-a-capriles-vote-of-less

    For what is worth, my own prediction as a former c21 and datanalisis employee is a tight ( or tighter than usual?) victory for chiabbe the hut (say 53 to 47). I very much hope i am wrong.

    • Boy, Smilde drives a Mack truck through Christiansen’s numbers there.

      I found it especially damning that Christiansen is going around saying the REP has grown by 1.5 million when in fact it’s grown by just 1.3 million. That’s a mistake a guy in a Panadería might get away with, but Luis Christiansen speaking at a formal event in New York!?!?? The guy’s entire professional reputation is supposed to rest on his mastery of this kind of detail, how can he misstate a key number like that?!

    • :)

      OK, this comment made me really happy: I feel like something I wrote actually made a difference to someone’s way of thinking, which doesn’t happen that often!

      19,000 voter sample? Definitely fake.

      Also, Datos does not make powerpoint slides. They disclose their findings in private, one-on-one meetings with clients that are prefaced by the signing of elaborate non-disclosure agreements. They’re widely assumed to use Barium Meal Tests to ensure they can track down the source of any leak. They are very effective at it.

      This thing in Notitarde is just poll-war stuff.

      • Oh, another tell:

        “Y se llevó a cabo en terminales de pasajeros interurbanos y rurales, centros comerciales populares, hospitales públicos, plazas, paradas de autobuses, estaciones de trenes y metro, universidades y vía pública, zonas industriales, comerciales y agrícolas. (Con información de la Encuesta Datos).”

        Sorry, proper polling is done in people’s homes. Everybody knows that.

  20. Se lo digo a mis amigos cada vez que hablamos de este tema. Una encuestadora cuya encuesta a tres semanas de las elecciones tenga un nivel de incertidumbre del 25% (2.7% de margen muestral + 23% de indecisos/nosaben/nocontestan) y que a 2 semanas de las elecciones ese error siga siendo 17% (2.7% + 15%) es un desastre monumental, una piratería sin precedentes. No digo que Datanálisis sean vendidos al gobierno pues un escenario posible es que estén escondiendo una “dura verdad” (aunque lo dudo). Mi crítica no es moral. Es técnica. Un encuestador que se precie de ser tal no puede anunciar 15% de indecisos. Si te sale ese resultado te callas y no publicas nada. Denota una falla estructural en el diseño de la encuesta. Recordemos que esta encuestadora fue la que dijo que no había ninguna posibilidad de que la oposición ganara el referendum constitucional…. y lo ganó.

  21. LVL is talking right now and saying the same thing that Grisanti just released in his latest Barclays report:
    - Gap between Chavez and Capriles has tightened, but Chavez has a 10 point.
    - If remaining undecided behave how poll suggests (72% for Capriles, 28% for Chavez), margin narrows to just 3 points (51.6% for Chavez, 48.4% for Capriles).
    - This would leave the race in a technical tie.

  22. I’m hearing slightly diffferent figures, but the same overall picture. Undecideds breaking 83.6% for Capriles, giving HCh 50.4% and HC 47.2%, a gap of 3.2%. Datanalisis joins the ‘technical tie’ group!!

    • Butt-covering, indeed. But from someone who was there, I hear he indicated that a close result was the most likely scenario, once you factor in how the so-called undecideds have been behaving. Seems like a significant shift of emphasis.

  23. Hemos hablado mucho de que los que no responden podrían estar mayoritariamente del lado de Capriles por aquello del factor miedo. ¿Alguien se ha paseado por la idea de que muchos de los que sí responden y están a favor de Chávez en las encuestas puedan ser empleados públicos o beneficiarios de misiones pro-Capriles que por (de nuevo) miedo no se atrevan a revelárselo a un tipo que le acaba de investigar hasta la marca de calzoncillos? Quizás ese 47% de Chávez ni siquiera llega a tal. En un país con miedo y con una población amenazada por su gobierno, las encuestas dicen poco.

  24. i’m sure datanálisis’ clients are pretty aware of what datanalisis’ numbers are saying. the “guabineo” is for the local media for obvious reasons… like some entries of this blog.

  25. OK guys, a lot of you are saying, the NS/NC are going to break this way or that way… you’re forgetting something, a non-negligible number of people is not going to vote. So you can’t take the total number of undecideds and split them up between the two candidates. What you should do is pick a % of abstention and re-normalize.

  26. Well, voter turnout for the past 5 elections has been: 2006 presidential: 74,50% , 2007 Constitutional Referendum: 55,89% , 2008 Local and Regional: 62,00% , 2009 Constitutional Amendment: 70,30% and 2010 Parliamentary: 66,45%. Not surprisingly, the Presidential election drew out the most voter turnout….

Comments are closed.