turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Discussions
- :
- x y equation determination for prediction purpose

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Aug 1, 2017 5:22 AM
(1167 views)

Dear all,

I need a help about to determine x-y relationship determination. I have 6 batches and I analyzed them at different days. (22 times). I try to graph x= timepoint y= measurement graph and then create the most suitable equation between them. Linear, qubic or etc... How can ı determine to most suitable equation to determine when timepoint x= 6 y=? value?

Is there anyone can help about this?

6 REPLIES

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Aug 1, 2017 6:01 AM
(1162 views)

JMP provides many platforms for fitting different kinds of models. Each provides a variety of statistics or criteria to help select the best model from the data. (This process assumes that you have not previously determined the model from a theory.) Linear models are a great place to start.

- Select Analyze > Fit Model.
- Select your dependent y data column and click Y.
- Now you have a choice:
- you can add terms until you decide that there is no bias or
- enter all possible terms and remove the terms you decide are not significant until bias appears.

- You can use hypothesis tests for null estimates for individual terms (t-tests in Parameter Estimates or F tests in Effect Tests).
- You can use information theoretic criteria such as AICc or BIC about the whole model.
- You can use either of these kinds of information in a manual selection process by
- clicking Run and working with Fit Least Squares platform or
- in an automated selection process by changing the personality from Standard Least Squares to Stepwise and then clicking Run.

You can then use a variety of ways to predict the response but the Prediction Profiler is probably the best.

If the linear model is in inadequate, then there are other models to choose from.

Please see **Help** > **Books** > **Fitting Linear Models**.

We also provide training courses that cover this subject, which is not easily taught through a simple discussion here.

Learn it once, use it forever!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Aug 1, 2017 6:25 AM
(1158 views)

Hi,

Could you please help me to determine at the attached jmp file to obtain the most suitable equation determination between Time/Result interaction?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Aug 1, 2017 6:52 AM
(1152 views)

(Please note that I included a picture of the initial Prediction Profiler in the first post by mistake. I replaced it with the updated Prediction Profiler that reflects the change in the Time to 6.)

Assuming that Batch is a random effect and ignoring it for the purpose of this model, I used Fit Curve to explore plausible candidate models. I included a few exponential decay models in addition to the polynomial models that you explored. Here is the ranking by AICc:

The best choice is the three-parameter exponential model and the quadratic is a close second choice. In fact, the first five models are within 4 AICc, so that they are all supported by this data. But we prefer parsimony over complexity. The first two choices seem reasonable given a plot of the result over time for all batches. The complexity of the other candidate models seems unwarranted.

The choice now, it seems to me, is whether you need extrapolation. The quadratic model is turning up while the exponential model is approaching an asymptote.

My preference is the exponential model in this case so I will use it to demonstrate the exploitation of the model. The same process works with any of these models.

- Click the red triangle next to select model (Exponential 3P in this example).
- Select
**Profiler**. - Change the Time(day) to the desired level of 6 by clicking on or dragging the Result trace or entering the value directly as the current value (red) in the scale.

The prediction is updated:

So your estimated Result at Time = 6 days is 98.46303, or 96.495 to 100.431 with 95% confidence.

Learn it once, use it forever!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Aug 1, 2017 10:58 AM
(1134 views)

**result **measurement over time would also be a factor in choosing which model is most suitable. So, if for instance the decrement possibility per unit time per unit measure of **result** is constant, the exponential function might apply.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Aug 3, 2017 1:28 AM
(1068 views)

When ı am selecting to most suitable curve (equation) for obtained results, I try all fit options and then look their Rsquare and RMSE values. If R2 value higher and RMSE value is lower it is most suitable equation for me. Therefore ı choose Quintic curve which is highlighted with yellow color on picture. But as ı understand from your comments, you decide the most suitable equation from AIC values? Am i right?

Is there a way deciding to most suitable curve for my data set? Which model is the most suit to my results?

Can you help about it

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Aug 3, 2017 7:06 AM
(1055 views)

Yes, you should **not** use R square or RMSE for model selection. Yes, you should either use a criterion such as AICc or use an *honest assessment* method like *cross-validation*.

Cross-validation uses two or three exclusive partitions of the data (*hold out sets*). The first set is used to *train* (fit) the model. The second set is used to *validate *the model selection. The optional third set is used to *test* the selected model. You can either designate the validation sets according to your own partition scheme or use *K-fold cross-validation*. You can take K-folds to the extreme and use *leave one out* validation.

I think that AICc is more than sufficient in your simple case of a single predictor (time).

So, yes, there are many ways to decide what is the most suitable curve for your data set. I also mentioned, and then another respondant re-iterated, that a theoretical model is usually the best when it is availble. The choice of the criterion for model selection and the selection of the best curve is up to you.

There is a reason that we have many criteria. There is a reason that JMP orders the candidates models by AICc.

Learn it once, use it forever!