TruthIsAll
09-16-2008, 07:57 AM
A Critique of Election Forecasting Models
TruthIsAll
http://www.geocities.com/electionmodel/ExposingElectionForecasters.htm
There are a variety of election forecasting models used in academia, the media and internet election sites. The corporate MSM (CNN, MSNBC, FOX, CBS, etc.) sponsors national polls to track the “horserace” and state polls to calculate the electoral vote.
Based on state polling data as of Sept.16, three well-known sites each have McCain in front: electoral-vote.com has McCain leading by 257-247 (34 tied). realclearpolitics has it at 227-207 (104 toss-up). The fivethirtyeight.com site also has McCain leading by 288-250. Splitting ties and tossups, electoral-vote.com has McCain by 274-264; realclearpolitics by 279-259.
The 2008 Election Model (EM) uses win probabilities based on projected state vote shares. Obama leads 285-253. Why the difference?
The EM uses Monte Carlo (MC) simulation method to calculate the probability of winning the electoral vote. Monte Carlo is widely used to analyze diverse risk-based models when an analytical solution is impractical or impossible. The EM is updated weekly based on the latest state and national polls. The model projects the popular and electoral vote, assuming both clean and fraudulent election scenarios. The EM allocates the electoral vote based on the state win probability in calculating a more realistic total Expected EV.
Corporate MSM pollsters and media pundits use state and national polling data. Electoral vote projections are misleading since they are calculated based on the latest state polls regardless of the spread; the state poll leader gets all of its electoral votes. This is statistically incorrect; they do not consider state win probabilities. And there is no adjustment for the allocation of undecided voters.
For example, assume that McCain leads by 51-49% in each of five states with a total of 100 electoral votes. Most models would assign the 100 EV to McCain. But Obama could easily win one or more of the states since his win probability is 31%. The 2008 Election Model would allocate 31 EV to Obama and 69 to McCain.
Bloggers also track state and national polls and do not adjust for undecided voters. A few use Monte Carlo simulation but the EV win probabilities and frequency distributions are NOT consistent with the polling data. Either the state win probabilities and/or the simulation algorithm is incorrect.
Academic regression models predict the popular vote but are run months prior to the election. They are typically based on economic and political factors rather than state or national polling data. They do not project the electoral vote. In 2004, virtually all of them forecast Bush to win by 5-10%. But since the election was stolen, the models had to be wrong – they didn’t factor election fraud as an independent variable in the regression. In fact, they never even mentioned the F-word in describing their methodologies.
There has been much discussion regarding the recent McCain “surge” in the national polls. Most national and state polls are sponsored by the corporate MSM. Gallup, Rasmussen and other national polls recently increased the Republican Party ID percentage weighting. This had the immediate effect of boosting McCain’s poll numbers. But there are 11 million more registered Democrats than Republicans. USA Today/Gallup changed the poll method from RV to LV right after the Republican convention. Party-ID weights were manipulated to favor McCain as well.
There is a consistent discrepancy between Registered Voter (RV) and Likely Voter (LV) Polls. The Democrats always do better in RV polls. No wonder: Since 1988, Democratic presidential candidates have won new voters by an average 14% margin.
The manipulation of polling weights is nothing new. Recall that the 2004 and 2006 Final National Exit Polls weightings were adjusted to match the recorded vote miscount. But all category cross-tabs had to be changed, not just Party ID. Of course, the Final Exit Poll (state and national) is always matched to the recorded vote even though it may be fraudulent – as it was in 2000, 2002, 2004 and 2006. This cannot be emphasized enough. Say it loud again and again.
In 2004, the 12:22am National Exit Poll (NEP) had a 38/35 Democrat/Republican Party ID mix. Kerry won the NEP by 51-48%.The weighting mix was changed to 37/37 in the Final NEP to force a match to the recorded vote miscount. Likewise, the Gore/Bush “Voted 2000” weights were changed from 39/41 to 37/43 in the Final NEP. Bush was the official winner by 50.7- 48.3% with 286 EV.
The final 2004 Election Model projection indicated that Kerry would win 337-201 EV with 51.8% of the 2-party vote. In their Jan. 2005 report, exit pollsters Edison-Mitofsky provided the average exit poll discrepancy for each state based on 1250 total precincts. Kerry won the unadjusted aggregate state exit poll vote share by 52.0-47.0% (2-party 52.5%) with 337 electoral votes - exactly matching the Election Model!
In the 2006 midterms, the 7pm NEP had a 39/35 Democratic/Republican weighting mix. The Democrats won the NEP by 55-43%. But the weights were changed to 38/36 in the Final NEP in order to match the 52-46% recorded vote; the Dem 12% margin was cut in half. Once again, the “Voted 2004” weights were transformed: from Bush/Kerry 47/45 at 7pm to 49/43 in the Final. The landslide was denied; 10-20 Dem seats were stolen.
The “dead heat” claimed by pollsters, bloggers and the media is a canard- unless they are factoring fraud into their models and not telling us. The media desperately wants a horserace. They fail to adjust the polls for undecided and newly registered voters. They avoid McCain’s gaffes, flip-flops and plagiarisms while he supports the most unpopular president in history.
TruthIsAll
http://www.geocities.com/electionmodel/ExposingElectionForecasters.htm
There are a variety of election forecasting models used in academia, the media and internet election sites. The corporate MSM (CNN, MSNBC, FOX, CBS, etc.) sponsors national polls to track the “horserace” and state polls to calculate the electoral vote.
Based on state polling data as of Sept.16, three well-known sites each have McCain in front: electoral-vote.com has McCain leading by 257-247 (34 tied). realclearpolitics has it at 227-207 (104 toss-up). The fivethirtyeight.com site also has McCain leading by 288-250. Splitting ties and tossups, electoral-vote.com has McCain by 274-264; realclearpolitics by 279-259.
The 2008 Election Model (EM) uses win probabilities based on projected state vote shares. Obama leads 285-253. Why the difference?
The EM uses Monte Carlo (MC) simulation method to calculate the probability of winning the electoral vote. Monte Carlo is widely used to analyze diverse risk-based models when an analytical solution is impractical or impossible. The EM is updated weekly based on the latest state and national polls. The model projects the popular and electoral vote, assuming both clean and fraudulent election scenarios. The EM allocates the electoral vote based on the state win probability in calculating a more realistic total Expected EV.
Corporate MSM pollsters and media pundits use state and national polling data. Electoral vote projections are misleading since they are calculated based on the latest state polls regardless of the spread; the state poll leader gets all of its electoral votes. This is statistically incorrect; they do not consider state win probabilities. And there is no adjustment for the allocation of undecided voters.
For example, assume that McCain leads by 51-49% in each of five states with a total of 100 electoral votes. Most models would assign the 100 EV to McCain. But Obama could easily win one or more of the states since his win probability is 31%. The 2008 Election Model would allocate 31 EV to Obama and 69 to McCain.
Bloggers also track state and national polls and do not adjust for undecided voters. A few use Monte Carlo simulation but the EV win probabilities and frequency distributions are NOT consistent with the polling data. Either the state win probabilities and/or the simulation algorithm is incorrect.
Academic regression models predict the popular vote but are run months prior to the election. They are typically based on economic and political factors rather than state or national polling data. They do not project the electoral vote. In 2004, virtually all of them forecast Bush to win by 5-10%. But since the election was stolen, the models had to be wrong – they didn’t factor election fraud as an independent variable in the regression. In fact, they never even mentioned the F-word in describing their methodologies.
There has been much discussion regarding the recent McCain “surge” in the national polls. Most national and state polls are sponsored by the corporate MSM. Gallup, Rasmussen and other national polls recently increased the Republican Party ID percentage weighting. This had the immediate effect of boosting McCain’s poll numbers. But there are 11 million more registered Democrats than Republicans. USA Today/Gallup changed the poll method from RV to LV right after the Republican convention. Party-ID weights were manipulated to favor McCain as well.
There is a consistent discrepancy between Registered Voter (RV) and Likely Voter (LV) Polls. The Democrats always do better in RV polls. No wonder: Since 1988, Democratic presidential candidates have won new voters by an average 14% margin.
The manipulation of polling weights is nothing new. Recall that the 2004 and 2006 Final National Exit Polls weightings were adjusted to match the recorded vote miscount. But all category cross-tabs had to be changed, not just Party ID. Of course, the Final Exit Poll (state and national) is always matched to the recorded vote even though it may be fraudulent – as it was in 2000, 2002, 2004 and 2006. This cannot be emphasized enough. Say it loud again and again.
In 2004, the 12:22am National Exit Poll (NEP) had a 38/35 Democrat/Republican Party ID mix. Kerry won the NEP by 51-48%.The weighting mix was changed to 37/37 in the Final NEP to force a match to the recorded vote miscount. Likewise, the Gore/Bush “Voted 2000” weights were changed from 39/41 to 37/43 in the Final NEP. Bush was the official winner by 50.7- 48.3% with 286 EV.
The final 2004 Election Model projection indicated that Kerry would win 337-201 EV with 51.8% of the 2-party vote. In their Jan. 2005 report, exit pollsters Edison-Mitofsky provided the average exit poll discrepancy for each state based on 1250 total precincts. Kerry won the unadjusted aggregate state exit poll vote share by 52.0-47.0% (2-party 52.5%) with 337 electoral votes - exactly matching the Election Model!
In the 2006 midterms, the 7pm NEP had a 39/35 Democratic/Republican weighting mix. The Democrats won the NEP by 55-43%. But the weights were changed to 38/36 in the Final NEP in order to match the 52-46% recorded vote; the Dem 12% margin was cut in half. Once again, the “Voted 2004” weights were transformed: from Bush/Kerry 47/45 at 7pm to 49/43 in the Final. The landslide was denied; 10-20 Dem seats were stolen.
The “dead heat” claimed by pollsters, bloggers and the media is a canard- unless they are factoring fraud into their models and not telling us. The media desperately wants a horserace. They fail to adjust the polls for undecided and newly registered voters. They avoid McCain’s gaffes, flip-flops and plagiarisms while he supports the most unpopular president in history.