From

Location

Message

Date Posted

Netherlands

I just took a good look at the 2013 OTT chart. From 900lbs to 2200lbs, each inch simply adds 8lbs. That's impossible. Coating a golf ball with one inch of lead will use a lot less lead than coating a beachball with one inch of lead.
This means smaller pumpkins will have way too high estimates and big pumpkins way too low. In other words, smaller pumpkins will go light and bigger ones heavy.
I'm wondering why this decision was made. IMHO the 2008 charts are more accurate using a more logarithmic approach. Perhaps somebody can explain.

8/19/2014 6:45:36 AM

Netherlands

Okay, the golfball/beachball example is not totally the same, but while making the marrow estimate charts a few years ago I noticed a logarithimc relationship and not a linear. I had the advantage of actually putting the marrows I measured on a scale while still growing.

8/19/2014 7:43:47 AM

Muddy Waters

I just got the 2013 Team Pumpkin Chart from Bart.I like it it makes my Tiny 1317 even heavier to the charts then 14%.Its awl Good! There just ball park guesstmates anyways.There will never be an exact science on this.its literally impossible.

Plus you have grower error in taping.I heard 1 guy say he likes to tape this way or that way,ignoring the system thats in place.That ruins the whole thing.lol

Adding a note here.As far as kin genetics go.I think only the top 50% of pumpkins should be used to consider the fruits average for if a kin has heavy offspring.The bottom 50% were most likely under fed & under watered.This would explain the smaller fruit & lighter weights per chart.So when researching genectics I would only look at the top half to make my decision.But do what ever floats your KIN!!

8/19/2014 7:46:33 AM

Western NY

I know your reasoning for thinking that way Brad, but for some reason it's a very linear relationship in that size range. It looks like it [theoretically, for now] picks up starting around the 500 OTT range.

Scott might be able to give a better technical reason as to why, but that's just how it is.

8/19/2014 8:19:39 AM

Netherlands

In theory OTT measures 4x height + 3x length + 3x width. So in theory tall pumpkins would tape more and therefore go lighter than low pumpkins.

EE= L+(2xH)
SS= W+(2xH)
Circ= (2xW)+(2xL)

For the marrow estimate chart I simply took EE x SS, but that also favours height, though not an issue in marrows as they are never tall.
I just don't see the linear relationship. Perhaps I should have paid better attention in math class...lol

8/19/2014 8:23:00 AM

Maine

where can I find the new OTT chart?

8/19/2014 2:12:42 PM

Princeton Kentucky

Page 6
, http://www.bigpumpkins.com/Attachments/Team-Pumpkin_2013_Enhanced_OTT_Chart.pdf

8/19/2014 3:46:41 PM

Willamette Valley, Oregon

Here's a link to the OTT Chart (AG on page 2):

http://greatpumpkincommonwealth.com/PDF/pumpkinmelon.pdf

8/19/2014 3:53:16 PM

Willamette Valley, Oregon

Interesting observation, Brad. Part of it is just due to rounding error. It's actually not exactly 8 lbs per inch, but in the 7.5 to 8.7 lb/inch range. If you look closely you'll see it starts with ~8 lb/inch then it includes some 9s and some 8s for awhile then changes to mostly 8's, so it's not linear there.

Beyond the rounding issue though, the 2013 GPC chart takes into account the fact that the pumpkins really aren't spherical, especially as they get bigger in OTT so the ball analogy doesn't hold especially when we get above 410" to 420" or so. You'll note that's where the old-style and new chart begin to differ noticeably. The new chart, on average, fits historical values better at higher OTTs. Last year the 2013 chart actually made pumpkins look a touch heavy at higher OTT's. While that wasn't my intent, it was probably nicer than going substantially light. We still don't have a lot of data in the really high OTT's so more changes may be in order at some point.

Here's a link to more info on why the new chart was needed:
http://greatpumpkincommonwealth.com/PDF/Holub_changes.pdf

8/19/2014 4:30:21 PM

Indiana

I have to agree with Brad here. The chart is off by over a standard deviation on pumpkins over 1500 pounds. Thats not any kind of measurement error. The estimate is clearly wrong. The whole point of an estimation chart is for it to be accurate. 45 data points over 1500 pounds is enough to get closer than averaging 6.4% heavy. I understand that data was after the fact, but the chart over 1500 pounds is light. Simple as that.

That said, I'm not trying to be critical here. I appreciate the VOLUNTARY work everyone has done on this.

Maybe I am weird, but it kind of torques me when the numbers are this far off. I don't need a feel good number so my pumpkin goes heavy. I want to know the comparative truth.

8/20/2014 4:21:29 PM

Eastern Pennsylvania

I'm an orange grower so for me life is simple. The comparative truth is the weight on the scale; I think the days of growing a 898 because of the % heavy are over. The days of growing a 898 lb pumpkin because it is orange and round are alive and well however. I think the point of the new chart was to reduce the % heavy of sub 1000 lb pumpkins, and I am sure the accuracy of the heavier ones will follow in the next few years.

8/20/2014 6:10:43 PM

Willamette Valley, Oregon

Tom -

Looking at a separation by pounds is a biased sample in itself.

Weightier (more massive by lbs) pumpkins tend to be "heavy" for their OTT. So when you look at the numbers that way you'll get a "heavier" average, because you SHOULD get a heavy average mathematically. Your 6.4% number is simply a mathematical artifact of the by-weight cut off. You can be forgiven for that mistake, though. I think that's why the old chart equation form hung on for so long.

A proper way to look at it is to sort the data by OTT and do some cut offs. Here are some results (Including the really "heavy" (by any chart) 2328 Meier),

the top 10 by OTT (>420in OTT) were +2.94% (heavy) on average.
Competing chart was -4.02% (light)

410" and above pumpkins were 1.92% (heavy) on average.
Competing chart was -1.91 (light)

It's not perfect for sure, but there you have some fair numbers to compare.

I'd encorage anyone with critism of the chart to contact me directly before making uninformed claims so I can fill you in outside of a public forum: bathabitat at gmail.com

8/20/2014 8:13:53 PM

Indiana

NO point in discussion here. You clearly think you have this right and have some sort of superiority complex. Nobody has been that condescending towards me in a loooong time.

Kudos on your hard work. I do respect that.

I will apologize for one thing. I was wrong about measurement error. People at certain weighoffs undershoot their pumpkins to make their pumpkins look better.

8/20/2014 11:10:16 PM

Willamette Valley, Oregon

Hey Tom,

I am frustrated, but I wasn't intending to be condescending. Sorry if you took it that way.

Understand that I put months of work in to developing the model behind the 2013 chart. Looking at residual plots, trying various equation forms, doing model fitting research, etc., before settling on the current model. Based on the 2011 and 2012 GPC data, the model Andy and I settled on using for the 2013 GPC chart was the best, unbiased fit to those data across the range of OTTs that we could develop.

It's frustrating and frankly offensive to me to have you post that the chart is off by x amount, when the analysis you use to make that conclusion is mathematically flawed. That's not an opinion, I can prove it mathematically, which is why, yes, I do think I'm right on this. Then you use your own incorrect analysis to say to me and the whole pumpkin growing community how "torqued" you are by the chart's inaccuracy. I think it's wholly appropriate for me to call B.S. on that. And I do.

The chart did yield slightly heavy pumpkins in the upper range of OTT. I showed above how much it was actually off, probably partly due to chance and partly due to some heavy genetics playing a role. I'm happy to discuss the reasons it was off the modest amount it actually was. But Tom, your approach to the question isn't mathematically valid. Presenting it as if it is and then insulting my work and me personally ("superiority complex") based on that is not acceptable.

8/21/2014 1:29:10 AM

Willamette Valley, Oregon

Here's why a weight-based cut off in OTT chart validation are incorrect and will include more "heavy-to-chart" pumpkins than "light-to-chart" pumpkins and lead to an average %-to-chart that is "heavy":

On a basic level it has to do with data censoring and selection bias. In order for a selection of regression data to be unbiased you have to select from the "x" or independent variable or "cause" variable (in our case OTT). When other criteria, such as actual weight, are used to select a cut off it can cause problems because actual weight is part of the calculation of our ultimate calculated output: %-to- chart).

A weight based cut-off necessarily EXCLUDES "light" pumpkins and INCLUDES "heavy pumpkins" thus resulting in an answer that appears "heavy". Consider this simplistic, but illustrative example:

Take 30 pumpkins, 10 each estimating 1398lbs, 1502lbs, and 1606 lbs:

OTT    OTT est Weight    Actual Weight    pct-to-chart
405    1398    1118    -20
405    1398    1258    -10
405    1398    1328    -5
405    1398    1370    -2
405    1398    1384    -1
405    1398    1412    1
405    1398    1426    2
405    1398    1468    5
405    1398    1538    10
405    1398    1678    20
417    1502    1202    -20
417    1502    1352    -10
417    1502    1427    -5
417    1502    1472    -2
417    1502    1487    -1
417    1502    1517    1
417    1502    1532    2
417    1502    1577    5
417    1502    1652    10
417    1502    1802    20
429    1606    1285    -20
429    1606    1445    -10
429    1606    1526    -5
429    1606    1574    -2
429    1606    1590    -1
429    1606    1622    1
429    1606    1638    2
429    1606    1686    5
429    1606    1767    10
429    1606    1927    20

That table might not show up well, but average actual weight is 1502lbs, average % chart is 0 which is what it should be in an unbiased chart.

8/21/2014 2:16:26 AM

Willamette Valley, Oregon

Now let's just look at those with actual weights greater than 1500:

429    1606    1927    20
417    1502    1802    20
429    1606    1767    10
429    1606    1686    5
405    1398    1678    20
417    1502    1652    10
429    1606    1638    2
429    1606    1622    1
429    1606    1590    -1
417    1502    1577    5
429    1606    1574    -2
405    1398    1538    10
417    1502    1532    2
429    1606    1526    -5
417    1502    1517    1

The average % chart is now 6.5% heavy, because of the biased selection procedure, which eliminated light pumpkins of higher OTTs and included heavy pumpkins of lower OTTs.

Finally lets do a selection based on OTT. Say 417 inches and above:

OTT    OTT est Weight    Actual Weight    pct-to-chart
417    1502    1202    -20
417    1502    1352    -10
417    1502    1427    -5
417    1502    1472    -2
417    1502    1487    -1
417    1502    1517    1
417    1502    1532    2
417    1502    1577    5
417    1502    1652    10
417    1502    1802    20
429    1606    1285    -20
429    1606    1445    -10
429    1606    1526    -5
429    1606    1574    -2
429    1606    1590    -1
429    1606    1622    1
429    1606    1638    2
429    1606    1686    5
429    1606    1767    10
429    1606    1927    20

Average OTT for the OTT based selection = 0%

Same thing happens if you use real data, but it's harder to keep track of what's going on.

8/21/2014 2:19:08 AM

Muddy Waters

It will never be perfect.Its just a ballpark.Lets have fun growing them & spend the time in a productive manner in the patch.I'm good with all the charts.It gives me multiple choices.You will never get it perfect.This is a battle that cant be one.

8/21/2014 5:40:13 AM

Muddy Waters

won!lol

8/21/2014 5:41:22 AM

Indiana

There is no doubt the math that you have provided all along is accurate. Unfortunately the assumptions are incorrect.

"Heavy and light" should only apply to individual data points, not entire subsets or ranges.

A weight estimation chart should always gravitate towards the average "0" whenever a subset between a certain range has enough data points to provide high confidence.

If only 3 of the top 40 pumpkins in the GPC list are "light" by the chart, that number should be a lot closer to 20 than 0. 40 is a large enough data set to make that assumption.

8/21/2014 2:59:10 PM

Willamette Valley, Oregon

I understand this is confusing. You're on the right track, Tom. The part I think you are missing (and I'm sure you're not the only one out there) is that the individual data points you mention needing to average 0 can only be OTT measurements (not actual weights), because OTT is the explanatory variable the model uses.

You can't look at the average %-to-chart of top WEIGHTS, because necessarily the %-to-chart is affected by the weight. (Actual weight is THE key component of the %-to-chart calculation, so using actual weight as a cut-off creates a self-fulfilling prophecy of heaviness.)

8/21/2014 4:23:54 PM

Willamette Valley, Oregon

While it has a gut feel of correctness, looking at just the upper range of actual weights does two main things in a wide swath around the cut-off point that make it a mathematically incorrect assessment of chart accuracy:

A) Light pumpkins, which the chart estimated to be a weight above the weight cut-off, fall below the weight cut off as a result of the %-chart-calculation. Thus they are improperly censored, or removed from the zero-tendency calculation. This drives the average %-to-chart up artificially. Example: A 1600lb estimated weight pumpkin comes in 10% light and weighs 1440 lbs. It's light, but gets omitted from the calculation because of its official weight. This pushes the average %-to-chart up artificially.

B) Heavy pumpkins, which the chart estimated to be a weight below the weight cut-off, end up above the weight cut-off as a result of the %-chart-calculation. Thus they are improperly included in the zero-tendency calculation. This also drives the average %-to-chart up artificially when using actual weight cut-offs. Example: a 1400 lb estimated weight pumpkin goes 10% heavy and has an actual weight of 1540 lbs and it gets included in the weight cut off. This also pushes the average %-to-chart up artificially.

It doesn't take too many omissions of light pumpkins and biased inclusion of heavy pumpkins to push an actual-weight-based cut-off approach into giving artificially high %-to-chart. The higher the worse this effect will be.

8/21/2014 4:24:19 PM

Willamette Valley, Oregon

This is the phenomenon I tried to demonstrate in the tables above. Please give that a close look.

So yes, at any given OTT length (not actual weight) the data should gravitate toward zero. And the new chart does just that across a range of OTTs, as it was designed to do (with some small error due to who-knows-what). It does not do that across upper ranges of weights, because that is not what it was designed or intended to do.

Furthermore, it is inappropriate to assess the model using actual-weight cut-offs. The reason we notice the apparent heaviness using weight cut-offs in the new chart is the bias in the old chart equations which masked it. When I first started looking into it, it was hard to get my mind around that fact.

Read this - look at the data table - Hopefully it will help too:
http://greatpumpkincommonwealth.com/PDF/Holub_changes.pdf

8/21/2014 4:25:56 PM

Willamette Valley, Oregon

It's worth noting that under 410" OTT or so, any of the recent charts will get you about as close as you could hope to be. So for the 95% of us in that OTT range there's really no reason to worry about which chart you use.

I do have my favorite though...any guesses? ha ha.

8/21/2014 4:39:02 PM

Willamette Valley, Oregon

I looked some more at the 2013 data. I made two graphs that should, I hope, clarify the central problems I have with using weight cut-offs.

The graphs should be up in the photo gallery soon:
http://www.bigpumpkins.com/gallery

Both graphs show all pumpkins greater than 392 inches, which was the OTT cut-off needed to encompass all 1500 lb pumpkins.

Several things to notice on the first graph (with measured weight on the horizontal-axis):

A) In order to encompass the 47 1500 lb pumpkins an OTT cut-off at 392 inches was needed. There were 131 pumpkins at or over 392 inches, so for an improper weight-based cut off, 84 pumpkins or 64%, almost 2/3rds, must be discarded! (I was even surprised by the scale of that number.) That's a big reason why the weight based assessment is such a wrong approach and the reason we see sooo many heavy pumpkins that way.

B) Within that OTT range there is a clear relationship between weight and %-to-chart. That's another reason why the weight-based approach is wrong: Weight and %-to-chart are correlated at the upper range!

In the second graph, with OTT on the horizontal axis, you can clearly see the biased inclusion and omission of data in the 1500 lb red group. Every open dot is one that should be included for the analysis to be fair and unbiased across the range of OTT where 1500lb pumpkins occurred.

Also note the lack of correlation between OTT and %-to-chart (as it should be).

Using instead a 417 OTT cut-off (approximately 1500 lbs), all open dots to the right of the cut off are points that should be included in a proper analysis. Any red-filled dots to the left of the 417 line should be excluded from a proper analysis; consistent with my explanation earlier today.

Anyway, I hope that helps clarify the problem. So as tempting as it is, please don’t fall into the trap of using a weight-based cut-off to assess an OTT-based chart.

8/21/2014 7:07:33 PM

Willamette Valley, Oregon

Here's the direct link to the first graph I discuss above:
http://www.bigpumpkins.com/displayphoto.asp?pid=7410&gid=1

And here's the second graph:
http://www.bigpumpkins.com/displayphoto.asp?pid=7411&gid=1

8/22/2014 12:29:23 AM

Everywhere there are growers

Almost everyone would agree that the old chat didn’t work to well and needed improvement. That was the reason Team-Pumpkin undertook the project of reviewing all the data and was first to publish an updated standard chart. Our new chart is the best fit to the data and is the best chart to date. We also introduced our Enhanced Chart which is the best tool available to evaluate pumpkins to see if they stand out from the pack and are heavier or lighter than normal. Without a lot of debate here I suggest you simply try our charts for yourself and see which works better. The charts are near the end of the article here on BP.

http://www.bigpumpkins.com/Attachments/Team-Pumpkin_2013_Enhanced_OTT_Chart.pdf

Team-Pumpkin

8/22/2014 6:28:55 AM

Willamette Valley, Oregon

Tom, there's another test for you. Run the Team-pumpkin chart through your incorrect weight-cut off approach. You'll find it is also heavy in that range (even thought the correct OTT approach shows it light.) The difference between the OTT and weight based cut-offs are about the same for both charts. Showing it is your calculation in error not either chart .

Actually I'll do it for you:

GPC chart: Top 50 by OTT = +1.92%, Top 50 by weight = +6.85, Bias from weight cut off = +4.93%
TeamP chart:Top 50 by OTT = -1.92% Top 50 by weight = +3.50, Bias from weight cut off = +5.42%

TeamP (and other older charts) are closer to 0 in the weight cut-off because the baseline they work from is "light". Don't let that deceive you.

8/22/2014 11:31:36 AM

Willamette Valley, Oregon

Bart, your unfounded claims make me smile.

Not that it has any relevance to chart quality, but the GPC chart was released at the Las Vegas meeting Feb 22-24, 2013. When was the TeamP chart released?

-------------

Best fit to data? Best chart to date? Those are some bold statements, Bart. I was happy to let it be with "both charts are fine for most of us", but since we're throwing around terms like "THE BEST"... Let's let people judge objectively for themselves looking at 2013 data:

Top 10 by OTT: GPC Chart +2.94% (without uow +0.87%) TeamP -4.02% (without uow -6.21%);
In terms of pounds of error on an 1800 lb pumpkin (Excluding uow):
GPC chart: On average pumpkins are 20 lbs heavy,
TeamP: On average pumpkins are 110 lbs light
Which is better there?

Top 50 by OTT: GPC Chart +1.92%, TeamP -1.91%
TIE! We are equally off from 0%, but don't forget the TeamP bias issues in the top 10.

Top 500 by OTT: GPC Chart +1.13%, TeamP +1.75%
Call it a tie, but one of us is closer to 0%.

In the 446 pumpkins around 1000 lbs (330 OTT to 390 OTT, ~800 to ~1275 lbs):
GPC Chart +0.8%, TeamP +2.5%
I ran some stats on these. We are both actually statisitically higher than the 0% it should be (p-values are 0.04 and <0.00001 respectively), but who's closer?

To be fair let's look below 200 OTT where the GPC chart suffers a little: GPC Chart -4.8%, TeamP +0.6%
So for little pumpkins TeamP is the one to use, although on a 150lb pumpkin 4.8% is 7.5 lbs, so I didn't worry too much about it.

For all 1354 pumpkins I have recorded from 2013: GPC Chart -0.16%, TeamP +1.30%
Prior to "the best" claims being made, I would have called that a tie, (except at the very high end of OTTs). but seriously, 0.16% off on average? It takes some cojones to complain about a result like that (Tom B) and even claim their chart is "the best" when it verifiably is not(Bart).

8/22/2014 1:22:53 PM

Willamette Valley, Oregon

---------------
Lastly...

Is the Team P enhanced chart a useful tool? Sure! It points out an important fact that for any individual pumpkin the potential error vs the chart is large. For both charts the Standard Deviation is ~9% over the whole range of OTTs. 68% of pumpkins should fall within one SD of the chart estimate. 95% of pumpkins fall within 2SD (i.e. +/- 18%). 99.7% fall within 3SD (i.e. +/- 27%).

8/22/2014 1:23:35 PM

Menomonie, WI (mail@gr8pumpkin.net)

It is an ESTIMATION chart...based on several peoples measurements. I would like to perform a Gage R and R. Have everyone measure the same pumpkin and put the measurements down. Until you have a 3D Scanner that measures it,your data will be all over

8/22/2014 2:07:57 PM

Eddyz@efirehose.net

Exactly Shannon, it's an estimate with thousands of variable and the outcome is just speculation. Is it winter already

8/22/2014 2:36:51 PM

Willamette Valley, Oregon

Short of doing a Gage R and R I have done a sensitivity analysis on the measurement error - I'll repost it here:

"An OTT measurement error of 3" too short would give a +1.7% heavy error in the weight estimate (at 420" OTT).

For a truly 0% pumpkin (one whose actual OTT indicates 0% to chart) to come in 10% heavy by mistake, the OTT would have to be off by around 18"! (so sytematically 6" off on each of the 3 OTT component measurements). "

8/22/2014 3:22:11 PM

Menomonie, WI (mail@gr8pumpkin.net)

Not saying the chart is not good...Just saying I would not bet my pay check on it. Vinyl Tapes being used to record measurements. No Stretch there.

8/22/2014 5:38:46 PM

Willamette Valley, Oregon

No worries. And I like the Gage R&R idea (That's gauge repeatability and reproducibility for folks.) If some sites have extra volunteers, have two crews measure all the pumpkins seperately and independently and see what that looks like. Some QA/QC procedures would put some questions to rest. And no I'm not volunteering for that. ha ha.

8/22/2014 6:30:47 PM

nOVA sCOTIA

thank f@#k someone mentioned the measuring tape.lol

8/22/2014 7:33:32 PM