Transcript 2016 JMP Summit E Poster No Videox

Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to
analyse data objectively from an agrochemical
rainfastness experiment with the use of JSL.
Experiment
Plants sprayed with
agrochemical product
One half of the plants
washed ~ 2 hours after
spraying and amount of
product on leaves
quantified – no rain value
Syngenta
•
Global Swiss agribusiness whose purpose is bringing plant
potential to life.
Rainfastness
•
Agrochemical products are generally sprayed onto leaves of
crops/weeds.
•
In heavy rainfall, some of the product may be washed off the leaf
surface, resulting in a loss of activity.
•
Rainfastness of products can be assessed in the lab by applying
the product to leaves followed by simulated rain and determining
the % product remaining after rainfall.
Second half of plants
subjected to rain then
washed and amount of
product on leaves
quantified – rain value
• To determine the uncertainty in % AI remaining after rainfall, the variability in both the rain and no rain
samples must be taken into account
• This is calculated using formula columns as it can’t be automatically calculated through JMP platforms
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Solution
Allow non-statisticians and JMP® novices to analyse data
objectively from an agrochemical rainfastness experiment.
Build a template data table with a series of scripts to visualise the
data, compare rainfastness of treatments to a control via a
Dunnett’s test.
The previous common processes were:
• to rely on the help of a statistician , or experienced person,
to analyse and interpret the data correctly.
• to ignore the variance associated with the rainfastness of
each treatment and simply compare the means, resulting in
incorrect conclusions.
• to ignore the variance associated with the ‘no rain’ samples
by dividing each ‘rain’ value by the average of the ‘no rain’
values. This resulted in smaller than actual confidence
intervals in the rainfastness of each treatment therefore
incorrect conclusions.
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
The user inputs raw data and treatment details.
The one click scripts allow the user to:
• Check that the data has been inputted correctly.
• Look for chemical degradation/carryover during the analytical
run.
• Check that the data satisfies the criteria to perform a Dunnett’s
test (equal variance).
• Display the results of a Dunnett’s test with a simple graphic that
demonstrates which treatments have significantly different
rainfastness to a control.
• Display a plot of rainfastness Vs. treatments with pooled
confidence intervals shown.
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 1: Paste in raw data
•
Paste sample name, analytical run order, and sample concentration data into the corresponding columns in the table.
Sample Run Order Sample [AI] /
for Analytical
Name ug/ml
21 1 t- a
88
38 1 t- b
97
137 1 t- c
88
189 1 t- d
85
97 1 t- e
80
79 1 t- f
90
151 1 t+ a
57
175 1 t+ b
58
168 1 t+ c
68
142 1 t+ d
67
Formula columns pull sample
information out of the sample name
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 2: Check that the data has been imported correctly
•
Run a script that plots a boxplot for each treatment/time point combination
•
Check the number of data points, y values (active ingredient concentrations) for rain Vs. no rain
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 2: Check that the data has been imported correctly
•
Run a script that plots a boxplot for each treatment/time point combination
•
Check the number of data points, y values (active ingredient concentrations) for rain Vs. no rain
Acknowledgements:
Niall Thomson,, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 3: Check for patterns with the analytical run order
•
Run a script that plots [AI] (proportional to product amount) Vs. analytical run order for each treatment/time point combination
•
•
Mostly positive correlations – indicative of carryover
Mostly negative correlations – indicative of chemical degradation over the analytical run
In this case, there is no
evidence of
carryover/degradation
over the run as there are
a mixture of positive and
negative slopes and none
of the slopes are
significant.
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 4: Fill Out Control Treatment and Groups for Dunnett’s Analysis
•
•
•
Under the ‘Control Treatment’ column, set the value for the control treatment, e.g. product with no additive, as yes. For other treatments, set the value as no.
Under the ‘Control Group’ column, set the value of all ‘no rain’ samples as yes, and the ‘rain’ samples as no
Under the ‘Compare to Control’ column, set the value of all ‘rain’ samples as yes, and the ‘no rain’ samples as no
For % product remaining
after rainfall
Control group = No rain
samples
Compare to control = rain
samples
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 4: Fill Out Control Treatment and Groups for Dunnett’s Analysis
•
Defining the control treatments and groups allows the rainfastness (% product remaining after rainfall) and confidence intervals for each treatment to be calculated, along with a
Dunnett’s test that tests the significance of the rainfastness of each treatment, compared to the control.
•
The Dunnett’s test is a multiple comparison test to compare a number of treatments individually to a control. This is analogous to multiple t tests, restricting the number of comparisons
to those including the control while maintaining an overall type 1 error rate (α* = α/K (0.05/number of comparisons), standard deviation = a pooled estimate across the treatments).
•
The test is performed via formula columns. It is not possible to use the fit model platform for this as the rainfastness is calculated using the rain and no rain values, so the variability
associated with both of these values has to be taken into account to manually generate the confidence intervals and pooled standard deviation.
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 5: Check that the data meets Dunnett’s test requirements
•
Run a script to check the heterogeneity of variance across treatments/timepoints
•
One of the assumptions of the Dunnett’s test is that the variance across treatments is equal.
•
If the standard deviation of each treatment/timepoint combination is not considered significantly different to the mean standard deviation (i.e. they lie within the shaded region of the
graph) then the Dunnett’s test is valid.
In this case the standard deviation in
product amount for each
treatment/timepoint combination is not
significantly different to the average
standard deviation so the Dunnett’s test is
valid.
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 6: Determine if the rainfastness of treatments are significantly different to the control
•
Run a script that plots rainfastness Vs. treatment.
•
The treatments with red points, above the critical value, have significantly different rainfastness to the control (Dunnett’s test, α=0.05). The treatments with green points, within the
critical range, are not found to have significant different rainfastness to the control in this test.
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath
Surviving the Rain: The Six-Step Programme for a
®
JMP
Novice
Stephanie Lucas, Senior Formulation Chemist, Syngenta, [email protected]
Stephen Pearson, Chemical Process Statistician, Syngenta, [email protected]
Alan Brown, Statistician, Syngenta, [email protected]
Challenge
Allow non-statisticians and JMP® novices to analyse data objectively from an agrochemical rainfastness experiment.
Step 6: Rainfastness Vs. Treatment Plot
Acknowledgements:
Niall Thomson, Dave Bartlett, Karen Meade, Anne Stalker, Mark Brittin, David Lomath