The recently finished National Marathon had 736 finishers in the
marathon, and provides splits at the half way point as well. Since
it's a modest size data set and readily available, I'll take the
opportunity to return to the positive vs. negative splits question
for marathon strategy. This race had qualifying time standards. This
will have likely skewed the racing towards people who have more
prior experience and who race faster, making it a good ground for
seeing what works well. The course was such that the first and second
halves were of approximately comparable difficulty. The first half
had the course's major hill, but was otherwise relatively flat.
The second half, net downhill, rolled its way along, never very long
or steep up or down hill. For this and other details, see
http://www.nationalmarathon.com/ My sanity check runner (guy I
know, and had talked to before the race about his plans) ran nearly
dead even splits.
The summary for those who don't like the gruesome mathematical details:
* People who ran very positive splits ('time in the bank' strategy) did
very poorly, none placing in the top 10% of their age group and few
making the top 50%.
* People who ran very negative splits did well, placing slightly more
than expected (albeit not statistically significant) in the top 10%, and
definitely significant at succeeding in placing in the top 50% and
avoiding bottom 10% finishes.
* People who placed well (top 10%) were extremely likely to have avoided
very positive splits, had splits much closer to even than the race median,
and were much more likely than racers overall to have run negative splits.
Gruesome mathematical details:
My transformed table is the seconds for the first half, seconds
for the second half, ratio of second/first (positive splits are greater
than one), and place as a percentage -- first of 100 runners is listed
as 0.01. Arguably this should be 0.00. The advantage of the computation
I used is that it will give the right number of runners finishing in
the top 10% of their age group -- anybody whose place is lower than 0.10.
Placement is by age group, not overall. Given that age does matter
to times, and that we're looking for good strategy -- regardless of
time for the race -- top age group competitors are a good place to
look. A 30-something guy can place relatively well overall without
having run a very good race for himself or his age group. But the
top 55 year old has to run well.
Consider the 10% (74) of runners who had the highest ratios (most
postive splits). If positive splits were a good idea, this group
should have done well (especially many top 10% finishes). If neutral,
then about 10% should be in the top 10%, 50% in the top 50%, etc.
Of the 74, 0 placed in the top 10%, and only 10 placed in the top 50%.
If it were an even neutral idea, those numbers should have been 7.4 and
37, respectively. (Due laughter at the 0.4 runner.) The differences
are significant at better than the 99% statistical confidence level.
25 of these runners placed in the bottom 10% of finishers in their
age group, statistically also very signficant. (Again more than 99%
confidence). So we can say with confidence that people who ran very
positive splits (the most even of these 74 took 21% longer on the
second half than the first, 1:54 and 2:18 for the halves, for instance)
had poorer than expected placement if the strategy were a neutral idea,
and much worse than expected if it were a good idea. 'time in the bank'
did not work.
Now consider the 10% of the runners who finished with the lowest ratios
(most negative splits). One might expect that these runners have
sacrificed some placement for the sake of their negative splits.
Instead, we find that 9 of them placed in the top 10% of their age
group (vs. the expected 7.4, not a statistically significant difference,
but it does not seem to have harmed their placement even at the upper
reaches), and 50 (vs. expected 37) placed in the top 50% of their
age group -- which _is_ signficant, at better than 99% confidence level.
Only 1 of these runners placed in the bottom 10%, statistically signficant
at, again, better than 99% confidence. So those who ran very negative
splits did no harm (as a group) to their top 10% finish chances, and greatly
improved their chances to finish in the top half, and to avoid the bottom
10%, respectively. The range of ratios for this group was from 84.4%
(splits of 2:10, 1:49) to 97.2% (1:58, 1:55).
So those who ran very positive finished poorly, those who ran very
negative did ok or well. But ... how about competing for the top
placements?
The 73 runners who finished in the top 10% of their age group (we
lost one to a small group) -- median ratio was 1.012 (1:32:56, 1:34:05),
versus the race median of 1.063. Lowest was 0.873 (near the lowest in
the entire race!) and highest was 1.184. Only 5 of these top finishers
were more than 10% positive split (ex, 2:00, 2:12), versus 251 runners
in the race overall. This is significant at a level far beyond 99%
(far enough that my tables don't go there -- z = 59). 31 of these runners
ran negative splits, versus 170 in the race overall. It is significant at
the 99% level that runners who ran negative splits were more likely
to place in the top 10%.
So runners who did well were especially likely to not run positive
splits, and were especially likely to run negative splits. Of the two,
avoiding very positive splits is especially well-marked, and 'very'
is an amount much less than many people have mentioned planning on running
their marathons. This is certainly not what the top finishers were
doing at this race.
Followup questions that'd be nice to ask include to check with the 5
who placed in the top 10% in spite of their very positive splits whether
this was a good race for them. Conversely, to follow up with the
31 who ran negative splits whether they wished they'd run the first
half faster.
The overall winner ran splits of 1:16:08, 1:14:47, to win by 3:37.
Second (part of a pack of 4 who passed half way in about 1:18:15)
followed that with a 1:16:16 second half, gaining 1:24 during the
second half over third place, who'd run 1:18:10, 1:18:39.
--
Robert Grumbine http://www.radix.net/~bobg/ Science faqs and amateur activities notes and links.
Sagredo (Galileo Galilei) "You present these recondite matters with too much
evidence and ease; this great facility makes them less appreciated than they
would be had they been presented in a more abstruse manner." Two New Sciences