Hi Josh and Brian,

I've made lots of remarks.  It took me a while, so please don't just
junk 'em.  On the other hand, I'm not a microbiologist, or any kind
of a biologist, so the style I prefer may not be appropriate for the
intended audience.  I won't be offended if you don't take my 
suggestions.   I hope you'll consider them though.

- Jason


Page/Paragraph Line/remark

1/1/10 "Research studies [indicate that] some of these intrinsic..."

1/last/last I might consider changing the last sentence to
"These models may be fit to observed data and used for statistical
inference about growth and death rates in microbial populations of
interest."

2/1/1-2 "One common disadvantage of these models is that they are not
sufficiently complex to accomodate populations exhibiting both growth and
decline over time."

2/1/5-6 "This approach ...  decline phase [] in the model."
2/2/2 "Although...systems, [there may be opportunity for useful application]
in other systems such as live animal production environments."

4/2/1 "A complete, crossed factorial ..."
4/2/5 "and the date [were] used to validata ..."

6/3/7 Can the equation [log N = Yhat + epsilon] be centered?
I have not seen a statistical model written this way before.  Typically,
Yhat indicates a prediction of random quantity from data.  In general,
hats convey dependence on observed data.  The model statement doesn't
need a hat.  Almost all
models such as yours, with additive errors, are written in the general
form  [ response = mean + error ].  Then the mean is estimated from data
(and that estimate is denoted using a hat.)
Just because Membre et all used the Yhat doesn't mean you have to.  
Also, I don't agree with the term "adjusted value."  I've attached a
few sentences giving an alternative presentation of the model.  In any
event, it might look nice to center model equations.

7/last and 8/first - 
I give an alternative presentation of the methods of analysis for the
D-values which you can read.  To compare the two columns in Table 3,
I can think of three things: 

1) computing their correlation coefficient, which is r=.999
2) checking that a the predicted D-values fall within 95% confidence
limits (or more appropriately, the even-wider 95% prediction limits)
based on the observed D-values
3) A paired t-test using the 9 differences.
I suggest these things in my presentation below, which says the same
things that you say on the end of page 7 and beginning of page 8.
I was uncertain about the multiple comparisons bit.  Did you use
Tukey?

"Analysis of variance of the 36 observed D-values
was carried out using the GLM procedure of SAS [citation].  This ANOVA
involved main and interaction effects for water activity and pH.  
Subsequent multiple comparisons among means for the nine treatment
combinations were carried out using Tukey's procedure. 
Agreement between the observed D-values and those predicted by the
model was investigated by computing their correlation coefficient,
by checking that the predicted values were within two standard errors
of the observed mean D-values and using a paired t-test of the 9
differences."

11/2/3 "... no significant differences were observed, [
indicating the plausibility] of the models ..."
11/3/lst two lines : Log N0 was [treated as a known parameter and 
assigned a Salmonella log population count of 7 at time 0.

12/2/14- 15 "Both exponential ..."  You already  did a good job of
explaining the general inactivation model on page 7 and
indicated that it would be used for populations not exhibiting
any growth.  Does that make these two lines redundant?  (I can see
why you've revisited them here on page 12, I'm just asking though.)

13/1/2 "...representing the  intercept and [slope for Aw], respectively."

13/1/4 "...representing the intercept, and slopes for pH and Aw, 
respectively."

13/2/4 "... are  evident in Figure 4 [;] Salmonella  populations 
declined..."

13/2/8-9 "... with b1 andb2 parameter estimate[s] of 0.02 and ..."

13/2/10 and 13/2/12.  It doesn't make sense to me report a 95%
confidence interval and to indicate "(P<0.05)".  Whether or not
the pvalue is less than .05 is indicated by whether or not the
hypothesized value for the parameter falls in the 95% confidence
interval.  I suggest removing both (P<0.05)s.

14/2/7 The statement that lower MSE indicates a more adequate
model is a little dicey.   It really only indicates the variability
of the response about the model for the mean.   You can have the
mean specified perfectly and still have variability about the
mean (and substantial MSE).   I recommend omitting "lower the 
MSE and the"

14/2/13 Should "S. Typhimurium" be italicized?

14/2/14 "This comparison [suggests] that the Churchill ..."

15/2/3 "...predicted  values of the pH 7 [] general inactivation..."

20/2/2 "... limitations and should not be [] relied upon [by
themselves]

In Table 3 on page 29, I'm not clear on what SEM(4)=2.42, near the
bottom, is meant to indicate.   Is this sqrt(MSE/3), the estimated
standard error for any of the 9 means, each of which is based on n=3
replicates?  If so, then it is estimated with 18 degrees of freedom,
which is the residual degrees of freedom if the entire experiment
was carried out: 

total df = (model df) + (error df)
26       =     8      +    18

Also, you might consider arranging these means and predicted values
this table as a 3X3 table, with pH as rows and Aw as columns, to
convey the response surface.  You could probably squeeze both 
observed mean D-values and model-predicted D-values into one
cell, as I've done in the attachment.