Libertarians can "expect" their 2024 aggregated national ballot return to fall within the range of 2,231,777 and 3,672,714.

This estimated range is based on realized returns, and does not account for any specific managerial measures in the interim designed to push the result above range. 2,231,777 is the linear estimate. 3,672,714 is the curvilinear estimate.

Lastly, it is largely unrealistic to use the 2016 Johnson returns (4,489,235 ballots) as a comparable upon which criticism of the Jorgensen campaign is laid. The 2016 Libertarian aggregate yield was very clearly an outlier, literally off the chart.

The Jorgensen 2020 campaign return (1,865,911) did undershoot the linear estimate (2,104,546) for this year's General by about 11%. That said however, the 2020 estimate necessarily included the massive 2016 outlier datum, which forced the trend estimates substantially higher for 2020.

In any case, the 2024 estimate range shows significant positive growth. Lastly, the results indicate, at least in the Great Plains and Inland Empire, that physical campaign events (e.g. bus tour) are somewhat correlated to larger vote yields.

This suggests that electorate movement does exist, at least marginally, and that voters are not necessarily positioned in the hardened duopoly dichotomy. In other words, support across significant "heartland markets" could potentially be increased with directed or "boutique" campaign efforts.

The Jorgensen campaign did very well, in my opinion.

Yes. The forecast is made upon observed data--i.e. it is what it is. As such, it is therefore entirely objective (excepting in certain instances where one might cleft the data due to outliers or a clear anomalous observation).

I use two datum points to establish a ranged estimate at any given ordinal point in time (i.e. an election year, which happens to be a sequential value). These estimates are nothing more than two "best fit" lines--one being linear, the other curvilinear.

Using both superimposed on the observed data provides a "base" range for the most likely total return...without trying to introduce subjective or hypothetical factors. I should note that whenever excluding an observation, the range should be run boths ways (with and without the cleft outlier). When asked for a point estimate, I use the average of the range estimate as explained above (i.e. I take its midpoint).

Because of its simplicity (and normally it is the more conservative estimate--normally), the linear function is perhaps preferred if you were to use only one of these two estimate functions. Again, this is not always the case, just normally the case. For example, the 2020 results from the national CP aggregate actually flipped these two estimate trend lines, making the curvilinear more conservative in a short period forecast (which is what you are seeking to determine).

You can verify the "flippage" (if that may be a word), because now, after the 2020 observation is included in the data, the percentage of the total national aggregate expected in future contests which field a national CP candidate is in negative trend.

As for the Libertarian estimate, it is quite clear that the 2016 result by candidate Gary Johnson is an anomalous observation. It more than doubled the curvilinear estimate for that cycle. This curvilinear forecast, pre-election, is based on the existing observed data.

So, I suppose the forecast is "biased" in the sense that the most immediate past election has a strong influence on the estimate, i.e. "more gravity". Indeed, by plugging in the most immediate past results, you are in effect changing the slopes of your forecast lines (they being weighted by the newest "observation").

Thus, the 2020 Jorgensen estimate (about 2.1 million off the linear) was pushed significantly higher by the massive 2016 Johnson outlier which "broke" the model. 2016 returned well over twice what is a typically "aggressive" curvilinear forecast; it was many deviations over the mean. These model breaches, incidentally, "fix themselves" with enough iterations of forward observed data (future election cycles).

If you want to compare the Jorgensen's result with other Libertarian runs, however, the question becomes how do you deal with the "gravity" of that massive 2016 return?

The way I did it was to simply extend the 2016 forecast (which is based on observations up to 2012) by an additional period and ignore the 2016 return temporarily. This provides a bracket on the Jorgensen result, without the heavy modification to the forecast slopes that would occur from including the immediate preceding 2016 datum.

Looked at this way, Jorgensen's 2020 results easily fell within the forecast range, and the 2020 result actually came close to breaking the high side (curvilinear) estimate. So, she did rather well. That said, to keep with objectivity, I model with the entire data set. That explains why I said Jorgensen had an 11% undershoot of the linear estimate...with the proviso on the 2016 returns. Lastly, because these models "self correct," my guess is that the 2024 estimate range (published above) will once again pretty much be spot on.

For completeness, here are the equations for the 2024 Libertarian vote estimate:

Linear: y = 192788x - 467255

Curvilinear: y = 43776e^0.3164x

The only caution I might add is that you would be advised to rein the curvilinear forecast at only one period forward. And definitely do not carry it more than two periods forward, or otherwise, it will probably yield an unrealistic estimate.

I did use the curvilinear function two periods forward (e.g. for 2016 and 2020) as a filter to permit a comparative analysis on Jorgensen's 2020 result--i.e. How well did she do, relatively? But that is a rare instance of me using the curvilinear out beyond one period.

More than you wanted to know. I am simply trying to be complete.

Libertarians can "expect" their 2024 aggregated national ballot return to fall within the range of 2,231,777 and 3,672,714.

ReplyDeleteThis estimated range is based on realized returns, and does not account for any specific managerial measures in the interim designed to push the result above range. 2,231,777 is the linear estimate. 3,672,714 is the curvilinear estimate.

Lastly, it is largely unrealistic to use the 2016 Johnson returns (4,489,235 ballots) as a comparable upon which criticism of the Jorgensen campaign is laid. The 2016 Libertarian aggregate yield was very clearly an outlier, literally off the chart.

The Jorgensen 2020 campaign return (1,865,911) did undershoot the linear estimate (2,104,546) for this year's General by about 11%. That said however, the 2020 estimate necessarily included the massive 2016 outlier datum, which forced the trend estimates substantially higher for 2020.

In any case, the 2024 estimate range shows significant positive growth. Lastly, the results indicate, at least in the Great Plains and Inland Empire, that physical campaign events (e.g. bus tour) are somewhat correlated to larger vote yields.

This suggests that electorate movement does exist, at least marginally, and that voters are not necessarily positioned in the hardened duopoly dichotomy. In other words, support across significant "heartland markets" could potentially be increased with directed or "boutique" campaign efforts.

The Jorgensen campaign did very well, in my opinion.

I agree with you, Floyd. Under the circumstances I believe she did exceptionally well.

ReplyDeleteI have a few interesting charts on Jorgensen's results for review. Would depend on Quirk posting them up, however.

ReplyDeleteFloyd, can you explanin how you come up with these figures? Thanks.

ReplyDeleteYes. The forecast is made upon observed data--i.e. it is what it is. As such, it is therefore entirely objective (excepting in certain instances where one might cleft the data due to outliers or a clear anomalous observation).

ReplyDeleteI use two datum points to establish a ranged estimate at any given ordinal point in time (i.e. an election year, which happens to be a sequential value). These estimates are nothing more than two "best fit" lines--one being linear, the other curvilinear.

Using both superimposed on the observed data provides a "base" range for the most likely total return...without trying to introduce subjective or hypothetical factors. I should note that whenever excluding an observation, the range should be run boths ways (with and without the cleft outlier). When asked for a point estimate, I use the average of the range estimate as explained above (i.e. I take its midpoint).

Because of its simplicity (and normally it is the more conservative estimate--normally), the linear function is perhaps preferred if you were to use only one of these two estimate functions. Again, this is not always the case, just normally the case. For example, the 2020 results from the national CP aggregate actually flipped these two estimate trend lines, making the curvilinear more conservative in a short period forecast (which is what you are seeking to determine).

You can verify the "flippage" (if that may be a word), because now, after the 2020 observation is included in the data, the percentage of the total national aggregate expected in future contests which field a national CP candidate is in negative trend.

As for the Libertarian estimate, it is quite clear that the 2016 result by candidate Gary Johnson is an anomalous observation. It more than doubled the curvilinear estimate for that cycle. This curvilinear forecast, pre-election, is based on the existing observed data.

ReplyDeleteSo, I suppose the forecast is "biased" in the sense that the most immediate past election has a strong influence on the estimate, i.e. "more gravity". Indeed, by plugging in the most immediate past results, you are in effect changing the slopes of your forecast lines (they being weighted by the newest "observation").

Thus, the 2020 Jorgensen estimate (about 2.1 million off the linear) was pushed significantly higher by the massive 2016 Johnson outlier which "broke" the model. 2016 returned well over twice what is a typically "aggressive" curvilinear forecast; it was many deviations over the mean. These model breaches, incidentally, "fix themselves" with enough iterations of forward observed data (future election cycles).

If you want to compare the Jorgensen's result with other Libertarian runs, however, the question becomes how do you deal with the "gravity" of that massive 2016 return?

The way I did it was to simply extend the 2016 forecast (which is based on observations up to 2012) by an additional period and ignore the 2016 return temporarily. This provides a bracket on the Jorgensen result, without the heavy modification to the forecast slopes that would occur from including the immediate preceding 2016 datum.

Looked at this way, Jorgensen's 2020 results easily fell within the forecast range, and the 2020 result actually came close to breaking the high side (curvilinear) estimate. So, she did rather well. That said, to keep with objectivity, I model with the entire data set. That explains why I said Jorgensen had an 11% undershoot of the linear estimate...with the proviso on the 2016 returns. Lastly, because these models "self correct," my guess is that the 2024 estimate range (published above) will once again pretty much be spot on.

For completeness, here are the equations for the 2024 Libertarian vote estimate:

ReplyDeleteLinear: y = 192788x - 467255

Curvilinear: y = 43776e^0.3164x

The only caution I might add is that you would be advised to rein the curvilinear forecast at only one period forward. And definitely do not carry it more than two periods forward, or otherwise, it will probably yield an unrealistic estimate.

I did use the curvilinear function two periods forward (e.g. for 2016 and 2020) as a filter to permit a comparative analysis on Jorgensen's 2020 result--i.e. How well did she do, relatively? But that is a rare instance of me using the curvilinear out beyond one period.

More than you wanted to know. I am simply trying to be complete.