Proc NLIN prediction limits for scored data (i.e., records with missing response values)

[BillT]

I am using NLIN to solve an 'errors in variables' model: Y = a + b*x  using iteratively weighted least squares regression where the weights are calculated as in York et al., 2004. The raw data for the response and explanatory variables are not available; we have the mean and the standard errors for each record. The equation used to calculate the weights includes the standard errors for Y and X, as also the  include the slope parameter (b). I am only using NLIN because of the need for iteration. The solution from NLIN has been verified.

I have two questions about scoring data that are not used to estimate the model. We have used the 'missing values trick' to produce predictions for records with missing values for the response, as well as confidence limits for the predictions. (https://blogs.sas.com/content/iml/2014/02/17/the-missing-value-trick-for-scoring-a-regression-model....) We have not been able to get SAS to produce prediction limits. In addition to my request for help in producing the prediction limits, I also would like to know how SAS produces the predictions and their confidence limits. The SAS documentation indicates the leverage and weights are used to produce the limits however, the scored data do not include weights or leverage values. https://documentation.sas.com/?docsetId=statug&docsetTarget=statug_nlin_details34.htm&docsetVersion=...

[PaigeMiller]

---

@BillT wrote:

We have not been able to get SAS to produce prediction limits.

---

Why not? What happens? Show us the code. Show us the LOG. Show us a portion of the data.

I also would like to know how SAS produces the predictions and their confidence limits.

Predictions are created using the fitted model equation. The confidence interval formulas are shown at the link you gave.

The SAS documentation indicates the leverage and weights are used to produce the limits however, the scored data do not include weights or leverage values.

If you don't have weights, then every observation gets a weight of 1. Every observation has a leverage, it is computed from the data.