I am very curious to hear our members responses to a paper from Mike Allen and Mike VandeHaar, Diet formulation for lactating cows: the good, the bad, and the ugly (2016 Southwest Nutrition Conference), that I uploaded to the Consultant Files section. It’s a critique of the ration balancing programs that the vast majority of the industry use today. I am especially interested to hear from consultants that have been in the industry for many years and have seen these programs evolve and can compare nutrition balancing success between these newest programs and some of the older systems that came before.
I copied some excerpts below but please read the paper in its entirety before sharing your opinions!
While these programs have greatly contributed to educating students and nutritionists about the complexities of the rumen, it is our view that they have not increased accuracy of nutrient supply to meet requirements beyond simpler models. Rather, because of the great complexity of the rumen, the many interactions among feeds, the animal, and microbes, as well as lack of knowledge, lack of accurate data, and known faults in model structure, these models likely reduce our ability to accurately supply nutrients to meet requirements compared with empirical models.
More complicated models are seductive for those who use them. These models provide a competitive marketing edge because of the common perception that precision and complication translate into greater accuracy. Of course, as mentioned earlier, this is rarely the case. Complicated models are often promoted by feed test laboratories because more feed analyses are required, which increases revenue, and by some companies because the models show that their products are required (even though direct experiments may demonstrate little effect). Complicated models also provide numerous opportunities for presentations and consulting by academics and others.
We view diet formulation programs incorporating mechanistic models as overly complicated and unnecessary for successful nutrition programs.
The same paper was published in Feedstuffs, September 5th 2016. It caught my attention then enough for me to write a letter to the editor. Pox on Feedstuffs as they never published my letter. Kudos to theCWTgroup platform that I now get to air my thoughts..
Feedstuffs Editors (9/27/2016):
My respect for the acumen around the understanding of dairy nutrition by Drs. Allen and Vandehaar triggered me to read their recent opinion in the September 5th issue of Feedstuffs titled Formulation Suggestions for Lactating Cow Diets. However, I disagree.
The complicated mechanistic models they refer to in the article have proven to be very useful to many of us in the trenches of dairy nutrition. A quote from statistician George E. Box I frequently use is “Essentially, all models are wrong, but some are useful”. This seems to apply quite well when considering the usefulness of dairy ration balancing software.
Although not exact, the formulation software tools available to dairy nutritionists have been useful in many areas…nonlinear least cost optimization, feed library organization, multi-pen diet management, DCAD tracking, herd to herd and pen to pen diet comparisons, animal wellness risks, environmental loading of nutrients from the dietary waste stream, and more. Allen and Vandehaar fail to mention these attributes. Instead they say, “we view the incorporation of mechanistic rumen models into practical diet formulation software as an exercise in futility”.
I believe it is reasonable to think that the minds of those having never earned a living by feeding dairy cows would differ from those that pull boots on every day and trust they’ll be compensated at the end of the month. Allen and Vandehaar mention, the success of the nutrition program is much more de pendent upon the nutritionist and the interaction and communication between the nutritionist and farm management, as well as personnel who feed, harvest the crops and purchase feeds. I agree, and said another way from many years ago; the eye of the master fattens the cattle. But complicated mechanistic models are very useful in along the way.
I think Daryl actually raises a bigger issue that I have with academia -or at least those that don’t have skin in the game – that if something does not fit their paradigm, they have the platform to pontificate without risks. I have seen this in the milk fatty acid area where Adam Locke refuses to acknowledge that their is any value to bulk tank milk fatty acids and with Miner institute and Cornell as it relates to NDFD and gut fill. Mike Allen and Mike VandeHaar make some good points about interactions and physiology that the current model have trouble dealing with, but to go out and bashing all models as having no value is wrong in my opinion. Thank you for adding this thought provoking post.
Having lurked more than reading and responding due to time restraints this on going battle is something that needs more opining although long emails are usually something that are usually better suited for strong coffee discussion or late night open dialog with some stronger beverages! However here it goes as a long email and when time to allow thought instead of reaction. The article is sent to me every time some one has read it or hears the presentation.
The discussion might be not two differing camps that are trying to decide which road to take at some determined fork in the industrial road, it might be better viewed as are we still better off using linear mathematics and then arguing over what set of constraints will give us the best outcomes for the CP, NFD, NFC, starch and NeL. Then we can discus what the outcomes are giving us. Then also maybe then we can even start the arguments over why the DMI is limited or if this is HOT or not. Many years ago a Cornell prof asked if ruminants even understand CP and NeL? So how do we try to understand glucose, amino acids and VFA? Are we now ready for non-linear mathematics? Most of us have used linear based mathematics and programming to help get us started years ago in understanding feeds and feeding. At one meeting within the last ten years Bill Chalupa did present a Pearson Square made in excel to formulate protein blends when CPM would not make a Composite or Mix but only an ingredient. Not even sure if linear program of Spartan makes an ingredient from a mix or not. Basically Bill did give way to using 6.0 CNCPS to properly hold a mix and make the proper calculation repeatedly. If we have the technology and practical application skills of understanding feeds with analysis and better computing do we not owe it to the industry to use this technology? Most non-linear mathematics makes it’s first steps with linear so maybe it is not an either or but a progressive step?
Attempting to find the BEST TOOL to use for the dairies that I get the chance to serve has made me a model user. Being solely responsible for the day to day nutrition for 15 to 25,000 cows on 5 or 6 dairies and several large feedyards full of dairy heifers (20,000) it is the best tool to use as the animals are observed, management worked with, health accessed, forages evaluated, profits examined monthly. In the last 10 years since leaving industry these herds have been given to me by attrition or last guy standing and nobody else wanted the headaches assignments, and not actively knocking on dairies doors. It has been a challenge and blessing both. Using it has opened up the opportunity of working with a CNCPS platform (Mike VA gave me a tour of it and then thanks to Sniffen and Cawood and too many beers in WI bar one late night!).
Last year at Tri-State both current CNCPS (3 platforms) and Spartan were discussed and then the rooms were assigned and break outs. This was the second time such format or pre-conference was offered. The model room was assigned as the original meeting room and the break out session was extremely full compared to the opening session. This is the same plan for the Southwest 2019 pre-conference. It appears the industry will make the decision.
Just as a side note, the previous reply was by David Weber. His display name was not set right which is now fixed.
Good question posted by Daryl. Good comments in response. Having worked in Michigan all of my career, I have interacted with the two Mikes for almost 30 years. I have learned much from them, especially Mike Allen as I worked for MSU Extension in the late 80’s and early 90’s. Those of us in Michigan who have a great amount of respect for these two, have also wrestled with their disagreement with some of the ration formulation programs that exist. I still remember sitting in the airport in Cedar Rapids, IA one afternoon, while Mike Allen explained how he used to think modeling was the future of nutrition and then became frustrated with the weaknesses of modeling as he saw them. And his frustration with the current group of ration software that uses models that he believes are inaccurate. He makes a very strong argument for his position. All that said, I would say two things.
One. I think we should ask Mike Van Amburgh to give a written response to the paper that Daryl references. I have had many personal conversations with Mike VA about this disagreement, and his answers always make sense to me. And I believe he has data to back up his position.
Two. I am old enough to remember Spartan 1, although I didn’t use it. I used Spartan 2 for over 10 years, switched to CPM around 2003, and switched to AMTS about 8 years ago. The last switch was made because CPM was no longer being supported and our team made the switch to new software together (and Spartan 3 would not work for us). I have achieved high levels of milk production with all 3 programs. However, I do believe AMTS (CNCPS) does help us account more accurately for what is happening in the cow and her environment and manage feed costs better than Spartan 2 did.
That’s all for now.
I heard Mike give a paper entitled Mind over Models at Tri-State in 2015. I thought he was spot on. I learned many years ago that people sell to people and the relationship of trust between client and consultant governs much of the success of our business. I also think the 2008 paper by Alex Bach makes the point very well when 47 different herds received the same ration for about a year and the range of milk production was 20-34 kg/day. That tells me the management role is primary. I am sure we have all experienced feeding similar diets to different herds with significantly different results. I have herds doing over 90 lbs on 2x with straightforward simple diets but with excellent management. We all know how much overcrowding affects herds from prefresh to milking groups. We have seen cows respond with more milk when forage quality improves from year to year. Mike emphasized in his paper that the cows will tell you how well the program is working. Cows biology follows a normal distribution and the model typically represents one cow in a herd of many. A few years ago I worked with Jim Beck attempting to develop a system that would classify cows by their insulin response to diets with 4 lbs more or less shelled corn. It was enlightening to see how about 20 % of the cows in the same herd increased in milk production when shelled corn dropped 4 lbs. My point is the normal distribution in biology is hard to capture and apply in a model. I have also observed in the field cows with a bulk tank average of 100 lbs in July with an average MUN of 5. I know the models have difficulty managing the nitrogen recycling ability of the cow. I think all of these and other issues affect what we do in our businesses. You need to find a program that fits your business.
Thanks everyone for your thoughts.
I have been thinking a lot about this topic the last several weeks and read some more papers related to it. I also uploaded the Mind Over Models paper that David Pullen mentioned. It goes into more detail about what Dr. Allen sees as the problems with the mechanistic models.
Most of the above posts do not specifically address Dr. Allen’s central argument; that mechanistic models are no more accurate than empirical models. I don’t think anyone would argue against the value of feed libraries, feed test downloads, feed mixes, DCAD computing, side-by-side ration comparisons, etc., but couldn’t you build all those things into a more simple model as well? Would you feel at a disadvantage if you were using an updated NRC 2001 model with all the newest data management features of NDS/AMTS?
What I keep coming back to when thinking about this argument, is that many of the very successful field nutritionists that I would consider mentors seemingly balance their rations the same way Dr. Allen is advocating for, probably without even thinking about it. While these nutritionists may use software programs that are based on mechanistic models, they do not use them they way I believe the software designers intended. They manually enter rations with the ingredient amounts they feel work well based on their experience and adjust the rations for CP, forage NDF, starch, NEl, and minerals. They do not worry much about the grams of MP or the lysine % of MP, etc.
And I don’t know if I can objectively say that I increase the success of the herds I work for with the time I spend agonizing over grams of AAs or uNDF or many of the other details. I definitely FEEL good when all the outputs are matched perfectly to the newest recommendations and I feel like I’ve done my part. But I am really starting to wonder if I could be spending that time on something more important or useful.
Thanks again for your opinions!