nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Error for empirical model estimates

abra

Hello dear community,

I am working with climate data in order to predict emissions of gasses from plant leaves. I created estimates for the amount of gasses (in gram) that was released over a season.

I performed several measurements in order to predict values for periods that were not measured.

I created two multiple regression models to predict for every hour the emissions (y2):

1) x1, x2, x3 to predict y1

2) y1, x4, x5, to predict y2

I predicted y1 and y2 values according to the estimates from the model. Then I summed up all the predicted values for y2 to get the estimate I needed.

What I lack is an error term for this estimate.

I though originally to get from jmp to predict the values of the error for each value by : fit model > Save columns > Std for predicted.

and then sum then up according to the formula:

n = 2500

Mutual SE = square root(se1^2+se2^2+......+se2500^2)

But the error barely grows as I create my cumulative estimate. so it does not seem as the right way

Any Ideas on how to approach this problem?

Thanks

Accepted answers

Peter_Bartell

Any number of modeling and simulation approaches are possible with JMP and JMP Pro. You don't say which version or product you are using. The beauty of JMP and JMP Pro is that you can visualize much of what you are trying to accomplish with some simple and elegant data displays. I suggest using the appropriate modeling and simulation platforms to give you the desired results. Then saving the model's prediction formula(s) out to the data table for use in the embedded JMP Profiler. If you have JMP Pro I also encourage you to utilize the Model Comparison platform to compare various models you have created. There is also an embedded Profiler inside the Model Comparison platform you may find useful too.

All comments

Peter_Bartell

Have you thought about a Monte Carlo simulation approach to solving for y2? I'm thinking what you are really after is not 'error' from a parameter estimate standard error point of view, but rather distributional characteristics of y1 and y2 for means, variances, shapes etc.?

abra

Thanks Peter for you answer,

I do not have experience (yet) with Monte Carlo.

I did not mention that I did this process several times, creating estimates for different plant species p1, p2... p6.

I would like to put error bars on these cumulative estimates in a graph, so I could show the reliability of the estimates and perform a contrast test between categories.

For these needs do you think this is the right method?

Peter_Bartell

abra

Only one last question,

How would you work when you have estimates that are the result of two model as above, where I first predict y1 and then use it to predict y2?

Since the prediction of error for y2 relies on two models.

Thanks