Assigning Column Filters to Variables and why am I getting an LS Mean of 0?Hi everyone!
I was wondering if someone could be able to help me. I am trying to analyze some data that is looking at inactivation of microorganisms over a period of time with different quality characteristics and with different sanitizers and different concentrations.
First, I am struggling to find the best column filter for my variables. I have two different sanitizers (A and 
, 3 levels of water characteristics (low, med, high) and time intervals at 0, 15, 30 and 60 seconds. Sanitizer A has 6 concentrations, while Sanitizer B has 4 concentrations. I have made the water characteristics into an ordinal numeric variables. Would the concentrations and time intervals be numeric continuous or ordinal?
Also, when I run a Standard Least Squares test some of my values are 0, and it tells me my model is missing an effect. Does anyone have any suggestions that may help me fix these issues?
Thank you so much!
Data Blending and CleanupRe: Assigning Column Filters to Variables and why am I getting an LS Mean of 0?Thank you so much I greatly appreciate it!
Re: Assigning Column Filters to Variables and why am I getting an LS Mean of 0?Hi,
I forgot to mention that one possible approach would be to run a Standard Least Square Model that incorporate Sanitizer, LOG (Concentration + 1), and COD Level and all interactions for each time point. At least at Time = 0:15, the distribution of the residual is close to a Normal distribution. Things get a bit trickier for Time > 0:15.
Best,
TSRe: Assigning Column Filters to Variables and why am I getting an LS Mean of 0?Hi,
Thanks for sharing the data.
A quick visual look at the relationship of your variable to the "Log (population)" outcome (my best guess on whether this is the right outcome) shows a clear effect of Sanitizer, Concentration, and Time on your outcome; the effect of COD Level is less clear but may exist.
The figure below where the replicates have been used to calculate the mean and standard deviation. Of note, the Time = 0:00 x COD Level = High shows a possible problem with you data: the point at concentration = 90 may not be correct.
Considering that your outcome shows a clear logistic relationship to Concentration and Time, you may need to seek more advanced expertise to correctly analyze your data: the standard Least Square Mean Model (Linear) is not appropriate because the distribution of the residuals deviates substantially from the Normal distribution (key assumption for this model).
I'll try to find a transformation of your data or a modeling approach that will meet your expectations. In the meantime, you may want to post a new inquiry in the board asking specifically about modeling biological response (i.e. Logistic) that do not meet the criteria for normally distributed residuals.
Sorry for not providing more concrete help.
Best,
TS
Re: Assigning Column Filters to Variables and why am I getting an LS Mean of 0?Hello,
My question is does sanitizer, concentration, time or water quality impact microbial inactivation? I am wondering what factor impacts microbial inactivation the most.
I assume as concentration and time increase, so will microbial inactivation, in addition to microbial inactivation decreasing as the water quality worsens.
I have included a snapshot of my data below.
I appreciate your help.
Re: Assigning Column Filters to Variables and why am I getting an LS Mean of 0?Hi,
You are asking a set of very broad questions while providing limited information. The first thing you need to figure out before performing any analysis in JMP (or any other analysis software) is what is your question and what are your assumptions. This has more to do with the technical nature of your experiment and less with the statistical tool you will eventually use. One way to approach this first step is to explore graphically your data including individual relationships of experimental factors with your data. Have you looked at the plot of microorganism inactivation data by all your variables? The nice thing in JMP Graph Builder is that you can easily nest several variables to figure out if there are gaps in your data.
The second general consideration you are asking about is the modeling format of your variables (I don't think you actually meant "filtering"). Here again, it depends on the question(s) you need to address. In general, if you have a continuous variable (e.g. time, concentration, physical measures), it is best to keep those as numerical continuous which should provide the best power in your model. The ordinal modeling is usually reserved for categorical variables that have a clear ranking but for which the spacing of each category may differ or be unclear (e.g. user preference, qualitative assessment).
Finally, would it be possible for you to share your data if these are not confidential or proprietary? This would help me and other on the board to figure out the best approach to your issue.
Best,
TS
