Showing posts with label attention projections. Show all posts
Showing posts with label attention projections. Show all posts

Saturday, March 14, 2009

Should Indiana Frontload in 2012? (Part Two)

The View from 2004 Continued

I left off the other day having predicted that Indiana would have received a much greater share (nearly 60 times greater) of candidate attention during the 2004 Democratic primaries and caucuses had the Hoosier state government opted to move their presidential primary from May to the earliest possible, non-exempt date (February 3). That Indiana would have gotten what amounted to about 5.25% of the total candidate attention -- in terms of both candidates visits and ad buys -- did not really leave us with any hard numbers in terms of what the state tangibly would have received. Instead of collectively then, let's look at the percentage of candidate ad buys and visits individually and parse this out a bit.

The model(s) here is (are) the same as it was in the previous post. All that is changing is the dependent variable. The result is that we'll construct two separate models (visits and ad buys) to get a clearer picture of what Indiana would have gotten out of an earlier presidential primary in 2004. Let's start with ad buys...

A Model Based on Candidate Ad Buys

Regression Analysis of State-level Ad Buy Shares (2004)
Variable
CoefficientIndiana Value
(actual/earlier)
Explanation of Measurement
Delegates
.026
1.87
Percentage of total Democratic delegates
Timing
.0001
106/15Number of days since first contest
Primary?
.941
1Dichotomous: 0 = Caucus
1 = Primary
Candidates
2.117***
2/7Number of candidates vying for nomination at time of contest
Events/Day
1.569***
0/7Number of other simultaneous events
Events/Week
-2.034***
0/7Number of other events in the same week (Wednesday-Tuesday)
Neighbors?
.018
0/0Percentage of neighboring states holding simultaneous events
Ad Buys (DV)
--
0
Average percentage of candidate ad buys
Constant
-4.646


R2 = .63 | n = 41 | Significance: *.05 **.01 ***<.01

In the 41 states where ad data was available for 2004, the number of candidates in the race during a particular contest and the number of events during a given Wednesday-Tuesday campaign week were significant factors (...as they were in the cumulative model). However, the number of simultaneous primaries and caucuses was also significant in this case. As was the case previously, though, the resultant relationship runs counter to what was hypothesized. It was expected that as the number of contests on any given day increases, the amount of attention -- in this case candidate ad buys -- would decrease. Again, this may have much to do with the high level of correlation between both "events" variables. To check this out, I ran the model twice more but without the events/day variable in one and the events/week variable in the other. In the events/day model, that variable loses its statistical significance, but the relationship with candidate ad buys is in the hypothesized direction. In the alternate model, events/week remains both statistically significant and in the hypothesized direction.

Before we get into interpreting what we see in the table above or the prediction, let me at least mention the model's fit. Performance-wise, this collection of variables explains nearly two-thirds of the variation in ad buy shares across states. That is an improvement over what was witnessed in the overall attention model. But effect of the statistically significant variables is largely the same (...save the events/day variable). A one candidate increase in the number of active candidates in the race creates an added 2.1% ad buy share for a state. Additionally, a one state increase in the number of states in a given Wednesday-Tuesday campaign week makes for a 2% decrease in a state's ad buy share. Beyond that, primary states got a bump of nearly 1% increase in their ad buy share over caucus states. Though that isn't a statistically significant finding it is substantively significant.

Fine, but what about Indiana? If the state had held its presidential primary on February 3, how much would the state's ad buy share have increased? After adjusting the timing (15 days after Iowa), the number of active candidates at the time of the contest (7), the number of other simultaneous events (7) and the number of other events in the same campaign week (7) variable values to reflect the that primary shift, Indiana would have significantly improved its stock. The Hoosier state had no ads bought/aired on its airwaves for its May primary (and thus a 0% share of the total ad buys). In February, however, the state's share of overall ad buys would have increased to nearly 8%, going from no ads to over 3000 ads aired. The other states holding delegate selection events on the same date (excluding Delaware and North Dakota because they had no data) averaged about 2700 ads aired. And while Indiana is predicted to have exceeded that number, the average of actual ads was weighted down by a relatively low number of ads aired in Missouri ahead of the Show-Me state's primary (see Gephardt discussion from the previous post).

Basically then, Indiana moves from getting nothing in May to garnering -- as the other states on the date had individually -- about half the ad buys as Iowa did to kick off the 2004 campaign.

A Model Based on Candidate Visits

And what about the other piece of the puzzle, candidate visits to the state? The results here are very similar to what we saw in the candidate ad buys model above.

Regression Analysis of State-level Visit Shares (2004)
Variable
CoefficientIndiana Value
(actual/earlier)
Explanation of Measurement
Delegates
.851
1.87
Percentage of total Democratic delegates
Timing
-.016
106/15Number of days since first contest
Primary?
.156
1Dichotomous: 0 = Caucus
1 = Primary
Candidates
1.414*
2/7Number of candidates vying for nomination at time of contest
Events/Day
.855
0/7Number of other simultaneous events
Events/Week
-1.627*
0/7Number of other events in the same week (Wednesday-Tuesday)
Neighbors?
.020
0/0Percentage of neighboring states holding simultaneous events
Visits (DV)
--
0.18
Average percentage of candidate visits
Constant
-1.303


R2 = .43 | n = 50 | Significance: *.05 **.01 ***<.01

Overall, the visits model isn't as good of a fit as the ads model. The included variables account for only 43% of the variation in the number of visits to a state (and that's despite the fact that there was visits data for all 50 states). However, the same basic group of variables was significant. Again, it is all about the number of candidates actively competing for the nomination at the time of a contest and the number of events in a week that matter at least statistically. The same phenomenon we saw above in terms of the two events variables is at play in this model as well. Namely, the events/week variable is significant and its relationship with visits is in the predicted direction, but events/day is neither significant nor in the hypothesized direction. Excluding each from the model has the same effect as well. Events/week doesn't change while events/day approaches statistical significance and has a negative effect on the number of visits to a state (as hypothesized).

Finally, a state's percentage of delegates also has a significant impact on a candidate deciding to touch down and actually campaign in a state. For each 1% increase in a state's share of delegates, the average state receives a bonus .85% of candidates visit shares. No, that doesn't seem like much, but the hypothetical difference between Indiana and, say, California would be 7% of the overall number of visits (based on delegates alone).

If Indiana would have moved its presidential primary to the earliest possible date in 2004, though, what would its share of candidate visits have been? The increase wasn't as great as it was in the case of ads, but Indiana's share of overall candidate visits increased nearly four and three-quarters percent by hypothetically shifting its presidential primary from May to February in 2004. At 4.69%, Indiana's predicted February share of candidate visits was 26 times greater than it was in actuality in May. A mere 6 candidate visits in May would have been nearly 160 visits had the Indiana primary been on February 3. The other seven states on February 3 averaged 57 visits, so the Indiana prediction greatly exceeds that average. The problem there is that the seven state average includes the depressed totals from Delaware and Missouri and the complete absence of visits to North Dakota for the state's caucuses. With those three states dropped from the average, the February 3 states averaged 88 candidate visits. Indiana, then, would have had a share of visits on par but shy of the number of visits South Carolina received on the same date.

Conclusion

In 2004, Indiana could have significantly improved the amount of attention it received from the candidates in the race had it shifted its presidential primary from May to February; going from 0 to 3051 ads aired and 6 to 158 candidate visits. But 2004 was unique in the opportunity it provided states in regard to moving delegate selection events in exchange for the spoils of the system. Only the Democratic Party had a contested nomination and their allowing for February contests opened the door for states to move, but only a handful of states took advantage of that rule change. However, those states on average increased their share of candidate attention. Yet that was probably a one and done proposition as many more states joined those seven at the front of the queue in 2008. The result was that those states saw a drop in the amount of attention each received. So, while Indiana could have significantly increased the attention the state received in 2004, that same increase would not have been available to the state in 2008 or in 2012 simply because, unless the rules change to regulate which states go when, there will be too much competition at the earliest allowable date (February 7, 2012).

The lesson? If the party rules change to allow for early dates, a state would be smart to move earlier rather than later to capture an increased share of candidate attention. Of course, Indiana was not in the most advantageous position prior to 2004. The Indiana House was controlled by Democrats who could potentially have been interested in moving the state's primary date for a competitive Democratic nomination (but never introduced a bill to do so). However, the state Senate was controlled by the Republicans. Had a bill to move the primary been introduced, the divided legislature could have proven a significant obstruction to such a move. It may, then, appear that Indiana missed an opportunity prior to 2004, regardless of legislative politics.

Up next? The 2000 primaries.


Recent Posts:
Michael Steele by the Numbers

GOP Temporary Delegate Selection Committee for 2012

Should Indiana Frontload in 2012? (Part One)

Thursday, March 12, 2009

Should Indiana Frontload in 2012? (Part One)

The View from 2004

A couple of weeks ago I laid out a simple model for examining the variation in the amount of attention (candidate visits and ad buys) states receive during any given presidential primary season. I further offered that one could use the information from this model to then predict the amount of attention a state would have gotten had it been in a different, more advantageous, calendar position. Let's reproduce the list of factors hypothesized to have an effect on the amount of attention a state would garner:
  • Delegates: As I have alluded to before, size maters. California is likely to get more attention from moving than Indiana.
  • Primary or caucus?: Despite all the chatter about caucuses in 2008, primaries still get the most attention from candidates.
  • Event Scheduling: This site is pretty much predicated on the idea that in the current system, earlier is better.
  • Number of candidates: Obviously, the greater the number of candidates in the nomination race at the time of a state's contest, the more attention that state is likely to get.
  • Number of simultaneous events: A crowded field of contests on any one day translates into candidate resources stretched thin. Look no further than Arkansas on this one.
  • Number of events in the same week: The reasoning above holds true here as well. If a state has a contest on the weekend following Super Tuesday, it may receive short shrift from the candidates than if it had not been as close to so many other contests.
  • Number of nearby states on the same date: Finally, resources are hypothetically more efficiently spent if a cluster of contests in neighboring states occur simultaneously. If John McCain is already in Missouri it is much easier (and more likely) to go campaign in nearby Oklahoma or Arkansas or Tennessee prior to February 5, 2008.
Since Indiana is in the process of looking into whether they want to form a committee to look into whether they should reposition their own presidential primary, I've been using the Hoosier state as my test case. [And incidentally, SCR 28 has passed the Senate by voice vote and has been introduced and read in the House as of March 10.] The question then is, "How much more attention would Indiana receive if it held its presidential on an earlier date?"

For the purposes of this exercise, we'll define earlier as the earliest date on which a state could hold its delegate selection event without sanction from the national party. And we'll be looking at this in terms of the 2004 primary calendar (Not to brag, but it is awfully nice to be able to reference all the primary calendars back to 1976 now.). I'll add in a 2000 projection later, but it is a bit messier with both parties having contested nominations. There are a couple of additional factors to consider. For now though, I'll focus on 2004, when only the Democratic nomination was at stake. In 2004, the earliest a non-Iowa/New Hampshire state could hold its contest was a week after the New Hampshire primary, February 3. Indeed, six states moved into the brave new world of February for the 2004 cycle, the cycle when the Democratic Party initially allowed for contests that early. In other words, there was some competition on that date but not anywhere close to the level of competition for attention on that same first Tuesday in February of 2008.

If Indiana had moved from the first week in May to the first week in February for the 2004 cycle, then, what would the Hoosier state have taken home? I looked at the descriptives of this recently and found that a handful of similarly-sized states to Indiana frontloaded in 2004 and gained as a result of the move. Tennessee and Wisconsin essentially went from nothing to around the amount of attention a state of their sizes would be expected to be if all other factors were equal. The other state, Missouri, unfortunately suffered because its favorite son, Dick Gephardt, was running and the contest had been pre-emptively ceded to the Congressman by the other candidates. [Yes, Gephardt had dropped out by this point, but the amount of attention the state got was far less than it would have been if Gephardt hadn't been in the race at all. He had only dropped out a couple of weeks prior and other states on February 3 -- especially South Carolina -- were getting much more attention.] Moving into February, then, had its advantages.

The Model

Before we look into the ramifications of Indiana having been the eighth state on February 3, 2004, let's look at how the underlying model performed.

Regression Analysis of State-level Attention Shares (2004)
Variable
CoefficientIndiana Value
(actual/earlier)
Explanation of Measurement
Delegates
.434
1.87
Percentage of total Democratic delegates
Timing
-.011
106/15Number of days since first contest
Primary?
.569
1Dichotomous: 0 = Caucus
1 = Primary
Candidates
1.496***
2/7Number of candidates vying for nomination at time of contest
Events/Day
.753
0/7Number of other simultaneous events
Events/Week
-1.366*
0/7Number of other events in the same week (Wednesday-Tuesday)
Neighbors?
.022
0/0Percentage of neighboring states holding simultaneous events
Attention (DV)
--
0.09
Average percentage of candidate visits and ad buys
Constant
-2.136


R2 = .47 | n = 50 | Significance: *.05 **.01 ***<.01

Looking at Indiana in particular, we're talking about a primary 106 days after the nomination race began -- well after it was over in fact -- that basically got nothing in terms of attention. Despite the fact that it is around the median for size and the fact that Indiana held the only event on its date or week, the Hoosier state primary just fell too far after the point at which the nomination had been decided to matter.

The Results

Across the board, though, what factors did matter?* I'm not terribly surprised that the percentage of delegates wasn't a significant variable. As I've said before, it just doesn't seem to matter in the context of frontloading. Timing didn't even matter, but that may have more to do with the fact that it is fairly highly correlated (>.8) with the number of candidates in the race at the time of the primary or caucus, which was a significant factor. Multicollinearity is a potential problem with the events/day and events/week variables. Obviously there is some amount of overlap between those two concepts, but the two are nearly perfectly correlated (>.9). That said, when events in a week is dropped, the simultaneous events on a date variable is still not significant. In events/day's absence, events/week remains significant. The oddity here is that with both are included in the model, they run in counter directions, which is not consistent with the hypothsized (the more events, the less attention). As it turns out, it is the significant variable (events/week) that runs in the proper direction. [Fortunately.] Together, these seven variables account for just shy of half of the variation in attention we see across the fifty states. Not a great fit, but not all that bad for a first pass.

The Prediction

With that baseline set, what level of attention can we predict Indiana would have gotten if it had shifted its presidential primary to February 3? We'll have to alter Indiana's numbers on events/day (7 events), events/week (7 events), percentage of neighbors going on the same date (0%) and the number of days after Iowa (15) to make this prediction. With those numbers imputed into the regression equation Indiana's predicted share of candidate attention rises from essentially nothing in reality to over five and a quarter percent had the Hoosier state's primary been held during the first week in February. That is consistent with the amount of attention Arizona actually got. Now, size doesn't matter in this model, but Arizona is a similarly-sized state to Indiana. Arizona did get the benefit of having a primary simultaneously with its eastern neighbor, New Mexico, which got a similar share of the total attention for the cycle.

Great, so, Indiana would have gotten about 60 times as much attention as it got in 2004 by moving from May to February, but how do we go about interpretting a share of attention. What does that mean in terms of the number of ads bought or the number of candidate visits to the state? I'm glad you asked. I'll pick up there tomorrow with part two.

*Yeah, but how do we go about reading those coefficients from the table above? We can see that the number of candidates in the race at the time of the contest and the number of events occurring in the same week as any given contest matter, but what do those numbers mean? In the case of the candidates variable, we can interpret that coefficient to mean that a one candidate increase translates to a 1.5% increase in the amount of attention a state receives. In a somewhat counteractive fashion, a one contest increase in the number of events in a given Wednesday-Tuesday week causes a 1.4% decrease in the share of attention a state garners. Of course the other variables play a role in determining this as well despite not being statistically significant. Substantively, both the percentage of delegates and primary/caucus distinction are significant, though the latter isn't as much as the extant literature might lead s to believe. In the case of the former, a one percent increase in the share of delegates a state had means a .4% increase in the amount of attention that state got.


Recent Posts:
WA-SoS Urges Steele to Back a Regional Primary System

2008 Electoral College by Congressional District

2008 Presidential Primary Calendar

Wednesday, February 25, 2009

If You Were Indiana, What Would You Do in 2012? A View from Similar States

Yesterday, I laid out a model for projecting how much attention the state of Indiana would have gotten had it held its presidential primary on an earlier date during previous cycles. Before we run the numbers, though, I thought it would be helpful to look at how other similar states have fared in moving their delegate selection events to earlier dates. Despite the fact that size just doesn't seem to matter (Neither my model here nor Ridout and Rottinghaus' model found the number of delegates a state has to be a significant factor in explaining the variation in a states decision to frontload or in the amount of attention said state receives.), my first inclination is to look at how similarly sized states fared after a frontloading move.

For our purposes, we'll use electoral votes as a proxy for size. Indiana has eleven electoral votes and fortunately, over the last several presidential election cycles there have been several similarly sized states that have made significant moves from one cycle to the next.


State Attention Changes (Delegates)
StateBeforeAfterAttention Change
Missouri
(11)
March 7, 2000
February 3, 2004
-0.37%
Tennessee
(11)
March 14, 2000February 10, 2004+1.32%
Wisconsin
(10)
April 4, 2000
February 17, 2004+2.39%

[NOTE: I should probably make at least some mention of the attention variable here. If you've had a chance to look at the Ridout and Rottinghaus paper cited above, the data they use to operationalize the concept is the number of candidate visits to and the number of ads run in a particular state. In the past I have used candidate spending and media coverage (a measure borrowed from Gurian and Haynes 1993) as a measure of this concept, but that series of measures has been compromised by both a change in what the FEC required candidates to file and by the proliferation of candidates opting out of the matching funds system.

As was my practice with the earlier set of data, I take the ads and visits data and convert it to percentages. Yes, that prevents us from saying, "Indiana would have gotten X more visits/ads had it moved its presidential primary Y number of of days/weeks forward." However, I find it powerful to look at this in terms of the share of attention each state got. "Could Indiana have increased its piece of the pie if it had been earlier?" If each state were created and treated equally then, we would expect them to each get about 2% of the total amount of attention (100%/50 states = 2% for each state). States, though, are not equal. Specifically, they are not equal in terms of size. A state like California, then, would be expected to net just more than ten percent of the total amount of candidate attention, while South Dakota would be expected to receive approximately one half of one percent. Indiana, in this case, is essentially right on average, expected to garner a hair more than 2%. From this, we can get a sense of whether a state has out- or under-performed based on what its expected share of the total amount of attention is.]

What, then, did Missouri, Tennessee and Wisconsin -- three states similar in size to Indiana -- gain from moving their presidential primaries to earlier dates between 2000 and 2004? In Missouri's case, not much. The Show-Me state actually lost ground, having gone from being lost in the Super Tuesday shuffle in 2000 to being stuck behind South Carolina and Arizona in 2004. The only reason I can think of to explain this is that the prevailing sentiment must have been that Missouri was going to be close in the general election anyway and that it was more important for the candidates to show strength in the South (in the Democrats' case) even if South Carolina was a done deal for November. First in the South status matters more than general election prospects, then.

While the Missouri case is somewhat puzzling, Tennessee and Wisconsin basically performed in the same way. Both went from essentially no attention in 2000 (both fell after Super Tuesday that year) to their expected levels of attention (give or take a few one-hundredths of a perentage point). What helped both in that regard was that both moved from inconsequential calendar positions to pre-Super Tuesday spots where they were not sharing the spotlight at all or with just one other state (Tennessee and Virginia went on the same date.). Overall though, this is a surprising finding (limited in scope though these cases may be). At least in terms of states around this size, we're talking about basically claiming a level of attention that should reasonably be expected and not out-performing that expectation. Basically, size may matter, but it is probably far less significant than how crowded the position to which a state of this size is moving.

Obviously size isn't all that matters, in fact across the two studies cited above, it has no real impact. But what are other means by which we can draw similarities between Indiana and other states? One thing that crossed my mind was that Nate Silver at FiveThirtyEight.com put together a nice chart of State Similarity Scores during the summer months of the 2008 campaign. [Sadly, the chart is no longer up in the post and I've tried it with three different browers. However, due to the magic of the internet, I was able to unearth the chart. Safari will allow you to right click the icon that is in the place of the image and save the original. I'm assuming that the reason it has disappeared is because the 1 GB Blogger allowance for images has been exceeded. But I could be wrong. Anyway, I'll reproduce the chart here, but only with the understanding that this is NOT my work. I am merely borrowing it for the purposes of this analysis.] Basically, Silver attempted to discover the percentage of similarity between states based on 19 dimensions.

State Similarity Scores
[Chart via FiveThirtyEight.com. Click image to enlarge and here to go to the original post there.]

Across those 19 dimensions, Indiana has the most in common with neighbor Ohio, Kansas and North Carolina. Kansas is a caucus state (with the exception of 1992) which makes it more difficult to equate with other primary states and North Carolina has been on the same date as Indiana for all but two cycles in the post-reform era (1976 and 1988). The Tarheel state got more than its expected share of attention in 1976 and matched its expectation in 1988. North Carolina is slightly larger than Indiana and serves as a nice baseline for comparison. Again, we see that sharing the spotlight on a particular date makes a difference (on its face at least). North Carolina had its late March date in 1976 all to itself (and was consequential in the GOP nomination race) but was part of the larger Southern Super Tuesday in 1988, a year when the state did better in terms of receiving attention than in the past, but not as good as it would have been had all the other Southern states not gone simultaneously.

The state that shares the most in common with Indiana, though, is Ohio. And the Buckeye state certainly offers a cautionary tale when it comes to the impact of delegate-richness on the level of attention a state receives. Despite being a bellwether during the general election, Ohio has consistently come in under its expected level of attention. Much of that has to do with the state having a presidential primary mired in either May or June for much of the post-reform era. When the state jumped to March in 1996 and ended up coinciding with the Illinois, Michigan and Wisconsin primaries, Ohio did gain attention from the candidates, but far less than would be expected for a state of its size and importance.* When Ohio moved again in 2000 (to Super Tuesday), the state actually lost ground. Again, the number of other states on the same date seems to matter.

Just looking through the data (without modeling and then predicting anything), then, size doesn't seem to matter that much. On top of that, if Indiana were smart, they would wait until the last possible moment in 2011 to decide on a date for its 2012 presidential primary (if they choose to go ahead with the move). In other words, the legislators in the Hooiser state would be wise to see where everyone else (or most everyone else) is going and then decide. The problems there, though, are twofold. The earliest allowed date, if party rules remain virtually unchanged, is going to be the first Tuesday in February, and as we've seen, it is going to be crowded. In other words, attention gains would be limited (The real gain would be in insuring that the state's voters have a say in who the party or parties' nominee(s) are going to be.). Relatedly, state legislators would also have to weigh the defiance option. If they go the Florida/Michigan route, Indiana risks losing delegates to the convention, but possibly gains much more attention in the process (especially if the "lose half their delegates" penalty on the GOP side remains the same). But the latter problem is more a problem for the national parties than it is for the states. The penalty just isn't strong enough to prevent states from defying party rules.

*This is even more curious since Ohio had aligned its primary with three other states in the region and those states had the date all to themselves. What didn't help was the fact that Dole's sweep through the remnants of the Southern Super Tuesday the week before forced his main rival, Steve Forbes, from the race, virtually assuring the senator of the GOP nomination


Recent Posts:
Indiana and 2012

Democratic Ulterior Motives for 2012 and Frontloading

Florida in 2012: Primary on the Move?