Presenting the Caracas Chronicles Legislative Elections Forecasting App, now in color!


Logo AppFive years ago, I got into a knock-down drag out fight with @puzkas (a.k.a. Eugenio Martínez, then El Universal’s elections guru) over the question of what must’ve seemed like a piece of arcane technical detail to many people: whether it’s possible to forecast the outcomes of a National Assembly election on the basis of “generic ballot” question (i.e., “do you intend to vote for the MUD or the PSUV candidate in your district?”)

Actually, calling it a knock-down drag out fight is embellishing it somewhat: mostly I berated him repeatedly in public and private for stating as a fact that no such forecast was possible, and he never actually responded. He just kept repeating as a mantra that A.N. elections were not national elections: the dynamics in each voting district (circuito) were different and therefore national trends told you exactly nothing about the overall outcome of the race.

I thought that position was non-sensical: of course district-level voting trends have some stable relationship with the national opinion climate. That relationship need not be linear, it need not be the same everywhere, and it need not be simple, but it was, in principle, calculable.

Then, @Econ_Vzla did the unexpected: he went ahead and calculated it. Using data from previous elections, he calculated how, on average, PSUV’s vote share varied in a given district as a response to changes in the national generic ballot question. It was a simple approach, simplistic even: but just because a model doesn’t have a thousand bells and whistles doesn’t mean it can’t get at an essential truth, and @Econ_Vzla’s model did just that.

@puzkas thought it beneath his dignity to respond, but still we wiped the floor with him over that fight. On the basis of its share of the nationall vote, our model predicted MUD would score 66 seats in the 2010 National Assembly. In fact, it scored 65. The model didn’t get every single seat right, but the ones it got wrong it got wrong in an unbiased way: it made as many mistakes in favor of the opposition as in favor of the government. As a tool for predicting overall seat shares, it was brilliant. Proof of concept was definitively established: you can predict A.N. seat distributions on the basis of national level opinion trends. That much we proved.

So naturally, we’re back at it this time around, though with a couple of important refinements.  Our 2015 model works not on the basis of the generic ballot question but of a more general poll question. Basically, we take as our single input the number of people saying they see the country’s situation as positive, and then we do the same thing we did in 2015: investigate the historic relationship in each district between the national trend and the propensity to vote for the government locally.

This year, though, we’ve decided to go one better: coding the model into a cool app that displays results visually, district by district, to give you a much more granular look of what the national opinion trend might mean locally. Other websites just tell you who the candidates are: Caracas Chronicles is sticking its neck out and telling you what results you might expect. The app’s brought to you courtesy of a whole bunch of sleepless nights and domino’s pizza consumed by our Managing Editor @emiduarte and developer @LuisErnestoZamb. It’s beautiful what they’ve done.

Go ahead and try it out, bearing in mind that on the basis of the latest Datanalisis poll the number you should be plugging into as an input is “12”. What comes out the other end when you do that is…a political earthquake of amazing proportions. But we’ll have lots more to write about that in the days to come.


  1. Awesome work, I’m playing with it right now. According to this tool, the MUD needs less than 12% of the electorate crazy enough to consider the situation “good” to get a two-thirds majority, and less than 23% to get three-fifths.

    • It’s probably best not to interpret the extreme cases too literally: what you see there is a product of some circuits having very low (and in a couple of cases, weirdly, negative) elasticity of PSUV-voting intention with respect to national level opinion trends.

      Where PSUV has piled up big enough majorities in the past, and those majorities don’t seem to respond very much to changes in the national MUD, the app doesn’t foresee those circuits going to MUD in any given scenario.

      You could interpret that as the model’s reading of what the absolute hard-core of impregnable PSUV fortresses…but there’s certainly room for debate about that. We’ll have more to say about that in the days to come.

      • Though there is something weird about that. I put 0%, just to see what happened, and it still shows us losing in 3 Zulia districts and 1 Carabobo district that we actually won in 2010. Shouldn’t those be a little more elastic?

    • When in doubt look, no further than massive ignorance and massive corruption.

      32 “ministerios”, close to 5 Million enchufados, plus all the “independent” or “undecided” “anti-imperio” “sosialistas rebolusionarioj” who still Adore their comandante Chavez, remember all the freebies, the false promises, and are still leeching profusely in various ways. That’s about 65% of the adult population right there.

  2. Godgiven’s face will turn as blue as the map of Monagas if he doesn’t even make it with such “posición salidora” in PSUV’s list.

  3. I’m not going to say the model is wrong. I do believe, however, that the premise of “how is the situation of the country right now” is flawed because we probably won’t know the real number on election day. But well. I also question the elasticity of some of the districts. For example, I enter “39%”, which yields more or less the results I’m kind of expecting, but with some strange results. It says the MUD is able to win BOTH of Monagas’s Districts, winning a district in Guárico and winning Vargas (places where we’ve never won) and losing in Amazonas, Nueva Esparta’s first district, and some of Zulia and Carabobo’s districts. I’m just wondering that.

    Puzkas is kind of right. It isn’t a national election, instead, it’s several elections held together, each with their own dynamics and characteristics. However, you would have to be dumb not to accept that the national situation of the country doesn’t affect it in any way. Wave elections do happen. Just look at the annihilation the Democratic party took last year at the hands of the Republicans. I do believe that a forecasting tool like this will be able to call many of the seats. However, like I said, several districts have their own dynamic, for example: Some of Táchira’s and Zulia’s districts are in that “Estado de Excepción” affair which limits the campaign oppo candidates can do (but probably won’t deter chavista candidates). So that could potentially change the outcome of those races.

    But other than that, it IS a very interesting tool and I have to commend the effort that went into it. We need more people doing stuff like this, so keep up the good work.

    • We’re much more confident that the model gets the overall result close-to-correct than that it gets any given seat right or wrong.

      We expect it to get given seats wrong, but we also expect that error to be unbiased: for every seat it gets wrong in PSUV’s favor it should get close to one seat wrong in MUD’s favor. But check back in on Dec. 7th!

      • In that case, the day before the election you’re going to have to say the number you expect to see if you were right! Looking forward to it.

        I do hope we win more seats than we are expecting. God knows we need a change.

  4. If the number from Datanalisis gets to 10 the model predicts qualified majority for the MUD! I can’t even imagine what the government would do in that case. I really hope the model is right.

    Question: does abstention count in the model at all? I remember many calculations for the presidential election where supposedly turn out numbers have a huge influence. Wondering if it might impact local races too.

  5. This is awesome! Congratulations to everyone involved.

    I agree with the premise that voting elasticities are informative “at the margin”. But what we’ve seen is a political debacle of industrial proportions. My smell test would be a comparison to the circuit-level polls that have been running around. In many of these, we see drastic swings towards the opposition in places like Guanare, Guarenas-Guatire and Southern Anzoátegui. However, these places are shown either light red or light blue in the model. Another topic to consider is the differential effects of the crisis in different areas, which might further worsen the government’s position in more rural areas where the crisis has hit worst. I’d think of these results as a lower bound for the opposition. But then again, the model also doesn’t account for what Lee Kwan Yew mentioned above.

    In any case, looking forward to the results! Congrats again.

  6. Why not use a more direct poll result as it was done in the original approach ? Unless you lack data , which it would be weird, it seems this approach adds a superfluous layer of analysis which will increment the uncertainty of the tool. Even more, given the almost universal consensus on the country situation (it sucks, horribly) , I expect the correlation between the variables: Perception of the situation of the country and voting intention to be more difficult to measure at this extreme politico-economical environment vis a vis a more normal one, like in USA for example. Remember that there is a not-negligible group who thinks the country is in the dump but will abstain, or, bless them, vote for the government .The reasons can be multiple: blaming the opposition, cynical economic opportunity, fear, etc.
    I am not saying the model will be particularly wrong, and by the way kudos for putting all this hard work, it just it seems strange to complicate things when a perfectly sound model was already tested and tried.

    • Using a measure of the political climate rather than using a direct poll of the president/psuv favorability was my preferred approach given the fact there is a huge structural change that makes hard to compare the series before and after Chavez passed

      • Fair enough, I stand corrected regarding the original method.Nevertheless it seems strange to use a proxy when you arguably can have a direct estimation of the voting preference via the regular polls. Probably the point is that you lack enough data on a by electoral district level, and that would be a shame, although I can understand it is difficult to have 87 different well conducted polls updated regularly.

  7. I don’t want to be pessimistic, but this time around, I think we can count on the government to pull out all the stops with their normal basket of electoral tricks, including intimidation, outright fraud, etc. Given the state of the country and the mood of the electorate, it is obvious how the elections SHOULD turn out. The burning question is how will the country react when they don’t turn out even close to something resembling fairness.

  8. Of course not. And whatever the final results, a percentile will be subverted somehow. But it seems clear that the majority will be MUD’s. The real question, the one PSUV must be asking themselves already, is what the exact objectives, along with their operational strategies, are that this new responsability for the MUD will allow and be fought for.

    I honestly think the electoral battle can be called a win already (hold ‘er steady! that’s all they need to do). A shame would be to stop there starting now. Even with a majority, which will likely not be allowed to be over 3/5 and definetly not over 2/3, this will be an uphill battle. I’d love to see what writers here think the strategic landscape will look like, beyond the ideological givens.

    • We had long, involved discussions about this. But @Econ_Vzla is pretty adamant that the model can’t reliably predict that, so it’s misleading to include it.

Leave a Reply