There is a great deal of variability in the shape of
individual lactation curves, but until now tools
for quantifying that variability have been very weak.
With MilkBot^{®} technology,
several new approaches to understanding changes in milk production
become possible.

The most immediate MilkBot® product is fitted
parameter values, generated by the fitting
engine. This in turn allows statistical
analysis of possible relationships between other variables and
parameters. Parameter values can be generated by us
from almost any data set containing milk weights and returned
to you for use in various ways. We can also
calculate residuals at each data point, and provide you with software
to calculate lactation curves from parameter values.

Are the cows up or down on milk?

Almost everything we try to control in managing a dairy herd,
including health, feeding, environment, breeding, and milking, will
influence daily milk production. Quantifying changes in milk
production is complicated by large variability which can easily hide an
important economic effect in random noise. For example, we
might make a ration change which costs $0.25 per cow per day.
Did milk production increase enough to recoup that cost? Is
it a profitable decision?

This type of question is extremely common, in management of individual
herds as well as in dairy research. Because of the nature of
tests used in statistical inference, our ability to make a
statistically valid conclusion in questions of this sort is
proportional to the number of animals in the trial, the size of the
expected response, and the amount of random variability, or variability
that cannot be explained by a known valid model. Even without
formal training in statistical methods, common sense and experience
face the same limitations. In the case of the $0.25 feed
change, it is unlikely that we can reliably answer the question whether
or not it pays, because we won’t have a large enough trial to
detect the small expected milk increase which is likely to be obscured
by normal variability in daily milk production over the several days or
weeks that the change in ration will take to reach its full
effect. The economic consequences are not small,
though, because $0.25 per cow per day becomes $9,000 for a group of 100
cows over a year. Yes or no, it could be a costly guess.

This daily variability in milk production has several causes, including
cows entering and leaving the milking group, and the normal rise then
fall of daily production after calving, the “normal lactation
curve”. A standard technique for decreasing random
variability, and increasing the power of statistics, is to
develop and validate a model which can explain a portion of the
observed variation, MilkBot allows a very sophisticated
approach to this general problem set, significantly increasing the
precision with which we measure changes in milk production

It is probable that many factors influence scale, ramp,
and persistence
of individual lactations, including genetics, transition management,
feeding, and environment. MilkBot^{®}
allows us to use standard graphical and statistical tools to study
those associations. For example, the two charts below
show scale and persistence for
about six million lactations, averaged by previous-days-dry
(pdd) for the previous lactation. There are some complex
issues in interpreting this data, but it is clear that MilkBot^{®}
has been able to pull out patterns that suggest some interesting
hypotheses.