ASA Connect

Hi all,

I am faced with a challenge on how to get the information I want from the data . 

I have data from various in-house projects on experiments using fish with tank as the experimental unit. Tanks in each project has around 30 fish inside tanks used in each project may vary. In many projects I see one or two tanks,occasionally 3, with very low feed intake compared to the rest of tanks irrespective of experimental factors. I would like to run an statistical analysis to see if some tanks always have the lowest feed intake and should need control measures. To do that, I am thinking of ranking tanks under each project in the ascending order of feed intake and give a value of zero to the first three tanks. But I am not sure if this is a good enough approach. I would highly appreciate any help with my two questions.

1. Is there any better approach than this one to rank tanks with poor feed intake?
2. Which statistical analysis is best choice to analyse the data and show the probability of getting a poor performance by tank?.

This will enable me to identify if there is any physical factors like location, activity around the tank etc. that causes this difference in tanks and take measures to control it.

Thank you very much,

Nafiha

------------------------------
Nafiha Usman
Aquaculture Research Center.
Norway
------------------------------

​What are your independent variables?  It would help if you can compare the effects of various independent variables within tanks but I suspect that that's not possible.

------------------------------
Emil M Friedman, PhD
emilfriedman@gmail.com
http://www.statisticalconsulting.org
------------------------------

Original Message:
Sent: 08-09-2017 06:43
From: Nafiha Usman
Subject: Variations among experimental unit

Hi all,

I am faced with a challenge on how to get the information I want from the data .

I have data from various in-house projects on experiments using fish with tank as the experimental unit. Tanks in each project has around 30 fish inside tanks used in each project may vary. In many projects I see one or two tanks,occasionally 3, with very low feed intake compared to the rest of tanks irrespective of experimental factors. I would like to run an statistical analysis to see if some tanks always have the lowest feed intake and should need control measures. To do that, I am thinking of ranking tanks under each project in the ascending order of feed intake and give a value of zero to the first three tanks. But I am not sure if this is a good enough approach. I would highly appreciate any help with my two questions.

1. Is there any better approach than this one to rank tanks with poor feed intake?
2. Which statistical analysis is best choice to analyse the data and show the probability of getting a poor performance by tank?.

This will enable me to identify if there is any physical factors like location, activity around the tank etc. that causes this difference in tanks and take measures to control it.

Thank you very much,

Nafiha

------------------------------
Nafiha Usman
Aquaculture Research Center.
Norway
------------------------------

Hey Nafiha,

A few things:

1. Are you interested in knowing:
a) Whether *specific* tanks are underfed across all experiments, or
b) whether there are *some* tanks that are underfed in any given experiment?
In other words, are you trying to identify "repeat offenders across the experiments", or instead "outliers within each experiments"? Or (everyone's favorite goal) both?

2) Are you interested in:
a) Identifying that the underfed tanks have lower values from a ranking perspective, or
b) Quantifying by how much the underfed tanks are underfed (which ranks will overlook)?

Perhaps, for some clarity, a simple visualization will help (and maybe spur some extra comments). Something like a scatter plot with:
X-axis: Tank index
Y-axis: Feed amount
Scatter Plot Data: X vs Y, with a different color for each experiment.

Since you mentioned controlling for independent factors, perhaps you could then repeat the plot above with:
Y-axis: E[Feed amount | Predictors Variables] ..... aka Yhat...... (estimated by whichever model you're using)
and also
Y-axis: E[Feed amount | Predictors Variables] - Feed amount .... aka the residuals...... (estimated by whichever model you're using)

Also a quick note on whichever model (or experimental design) you're using to control for variation between tanks would be helpful.

With that, perhaps the next step will be all the clearer.

Hope this helps! Feel free to follow-up if you have any questions!

------------------------------
Glen Wright Colopy
DPhil Student
University of Oxford
------------------------------

Original Message:
Sent: 08-09-2017 06:43
From: Nafiha Usman
Subject: Variations among experimental unit

Hi all,

I am faced with a challenge on how to get the information I want from the data .

I have data from various in-house projects on experiments using fish with tank as the experimental unit. Tanks in each project has around 30 fish inside tanks used in each project may vary. In many projects I see one or two tanks,occasionally 3, with very low feed intake compared to the rest of tanks irrespective of experimental factors. I would like to run an statistical analysis to see if some tanks always have the lowest feed intake and should need control measures. To do that, I am thinking of ranking tanks under each project in the ascending order of feed intake and give a value of zero to the first three tanks. But I am not sure if this is a good enough approach. I would highly appreciate any help with my two questions.

1. Is there any better approach than this one to rank tanks with poor feed intake?
2. Which statistical analysis is best choice to analyse the data and show the probability of getting a poor performance by tank?.

This will enable me to identify if there is any physical factors like location, activity around the tank etc. that causes this difference in tanks and take measures to control it.

Thank you very much,

Nafiha

------------------------------
Nafiha Usman
Aquaculture Research Center.
Norway
------------------------------

Thank you both for your kind responses.

@Emil Friedman - You guessed right.  I do not have measured value for  independent variables within a tank like water flow, temperature, oxygen etc.

@Glen Wright Colopy - I am interested in knowing  
1.  Whether *specific* tanks are underfed across all experiments.
2. Identifying that the underfed tanks have lower values from a ranking perspective was my initial interest.  After reading your reply, the option of quantifying by how much the underfed tanks are underfed may also help me to identify if the problem is with the feeding system. I will give it a try.

Based on visualizations of ranks, I could see a couple of tanks had the lowest feed intake in about 50% or more of the experiments (have about hundred experiments and two hundred tanks in total). You suggestions gives me some more ideas to play around with the data. Thank you!.



------------------------------
Nafiha Usman
Researcher
Skretting ARC
------------------------------

Original Message:
Sent: 08-10-2017 10:56
From: Glen Colopy
Subject: Variations among experimental unit

Hey Nafiha,

A few things:

1. Are you interested in knowing:
a) Whether *specific* tanks are underfed across all experiments, or
b) whether there are *some* tanks that are underfed in any given experiment?
In other words, are you trying to identify "repeat offenders across the experiments", or instead "outliers within each experiments"? Or (everyone's favorite goal) both?

2) Are you interested in:
a) Identifying that the underfed tanks have lower values from a ranking perspective, or
b) Quantifying by how much the underfed tanks are underfed (which ranks will overlook)?

Perhaps, for some clarity, a simple visualization will help (and maybe spur some extra comments). Something like a scatter plot with:
X-axis: Tank index
Y-axis: Feed amount
Scatter Plot Data: X vs Y, with a different color for each experiment.

Since you mentioned controlling for independent factors, perhaps you could then repeat the plot above with:
Y-axis: E[Feed amount | Predictors Variables] ..... aka Yhat...... (estimated by whichever model you're using)
and also
Y-axis: E[Feed amount | Predictors Variables] - Feed amount .... aka the residuals...... (estimated by whichever model you're using)

Also a quick note on whichever model (or experimental design) you're using to control for variation between tanks would be helpful.

With that, perhaps the next step will be all the clearer.

Hope this helps! Feel free to follow-up if you have any questions!

------------------------------
Glen Wright Colopy
DPhil Student
University of Oxford

Original Message:
Sent: 08-09-2017 06:43
From: Nafiha Usman
Subject: Variations among experimental unit

Hi all,

I am faced with a challenge on how to get the information I want from the data .

I have data from various in-house projects on experiments using fish with tank as the experimental unit. Tanks in each project has around 30 fish inside tanks used in each project may vary. In many projects I see one or two tanks,occasionally 3, with very low feed intake compared to the rest of tanks irrespective of experimental factors. I would like to run an statistical analysis to see if some tanks always have the lowest feed intake and should need control measures. To do that, I am thinking of ranking tanks under each project in the ascending order of feed intake and give a value of zero to the first three tanks. But I am not sure if this is a good enough approach. I would highly appreciate any help with my two questions.

1. Is there any better approach than this one to rank tanks with poor feed intake?
2. Which statistical analysis is best choice to analyse the data and show the probability of getting a poor performance by tank?.

This will enable me to identify if there is any physical factors like location, activity around the tank etc. that causes this difference in tanks and take measures to control it.

Thank you very much,

Nafiha

------------------------------
Nafiha Usman
Aquaculture Research Center.
Norway
------------------------------

​You might also want to plot intake versus time for each tank.  Depending on the number of tanks you might be able to overlay and symbol/color code them.

------------------------------
Emil M Friedman, PhD
emilfriedman@gmail.com
http://www.statisticalconsulting.org
------------------------------

Original Message:
Sent: 08-11-2017 08:28
From: Nafiha Usman
Subject: Variations among experimental unit

Thank you both for your kind responses.

@Emil M. Friedman - You guessed right.  I do not have measured value for  independent variables within a tank like water flow, temperature, oxygen etc.

@Glen Wright Colopy - I am interested in knowing
1.  Whether *specific* tanks are underfed across all experiments.
2. Identifying that the underfed tanks have lower values from a ranking perspective was my initial interest.  After reading your reply, the option of quantifying by how much the underfed tanks are underfed may also help me to identify if the problem is with the feeding system. I will give it a try.

Based on visualizations of ranks, I could see a couple of tanks had the lowest feed intake in about 50% or more of the experiments (have about hundred experiments and two hundred tanks in total). You suggestions gives me some more ideas to play around with the data. Thank you!.



------------------------------
Nafiha Usman
Researcher
Skretting ARC

Original Message:
Sent: 08-10-2017 10:56
From: Glen Colopy
Subject: Variations among experimental unit

Hey Nafiha,

A few things:

1. Are you interested in knowing:
a) Whether *specific* tanks are underfed across all experiments, or
b) whether there are *some* tanks that are underfed in any given experiment?
In other words, are you trying to identify "repeat offenders across the experiments", or instead "outliers within each experiments"? Or (everyone's favorite goal) both?

2) Are you interested in:
a) Identifying that the underfed tanks have lower values from a ranking perspective, or
b) Quantifying by how much the underfed tanks are underfed (which ranks will overlook)?

Perhaps, for some clarity, a simple visualization will help (and maybe spur some extra comments). Something like a scatter plot with:
X-axis: Tank index
Y-axis: Feed amount
Scatter Plot Data: X vs Y, with a different color for each experiment.

Since you mentioned controlling for independent factors, perhaps you could then repeat the plot above with:
Y-axis: E[Feed amount | Predictors Variables] ..... aka Yhat...... (estimated by whichever model you're using)
and also
Y-axis: E[Feed amount | Predictors Variables] - Feed amount .... aka the residuals...... (estimated by whichever model you're using)

Also a quick note on whichever model (or experimental design) you're using to control for variation between tanks would be helpful.

With that, perhaps the next step will be all the clearer.

Hope this helps! Feel free to follow-up if you have any questions!

------------------------------
Glen Wright Colopy
DPhil Student
University of Oxford

Original Message:
Sent: 08-09-2017 06:43
From: Nafiha Usman
Subject: Variations among experimental unit

Hi all,

I am faced with a challenge on how to get the information I want from the data .

I have data from various in-house projects on experiments using fish with tank as the experimental unit. Tanks in each project has around 30 fish inside tanks used in each project may vary. In many projects I see one or two tanks,occasionally 3, with very low feed intake compared to the rest of tanks irrespective of experimental factors. I would like to run an statistical analysis to see if some tanks always have the lowest feed intake and should need control measures. To do that, I am thinking of ranking tanks under each project in the ascending order of feed intake and give a value of zero to the first three tanks. But I am not sure if this is a good enough approach. I would highly appreciate any help with my two questions.

1. Is there any better approach than this one to rank tanks with poor feed intake?
2. Which statistical analysis is best choice to analyse the data and show the probability of getting a poor performance by tank?.

This will enable me to identify if there is any physical factors like location, activity around the tank etc. that causes this difference in tanks and take measures to control it.

Thank you very much,

Nafiha

------------------------------
Nafiha Usman
Aquaculture Research Center.
Norway
------------------------------