Understanding the prediction graph

[#1]

How to understand and what are all information we need to take it from the attached Actual vs Predicted graph as attached here

[#2 · Accepted solution]

Hi @Sankaramuthu , 

I think there needs to be more background information to make a thorough assessment on the actual by predicted plots. Especially, what model effects you considered (just main, also interactions or non-linear/quadratic effects, ...),  In general you will have a model that predicts the actual data well when the points are close around the line from the left bottom to the right top. 

AddendumHeight_mm and Teeth height_mm seem to fulfill this (though it would be important to know what variance you expect to be ok, aka if a deviation of .01 would be critical this might not be good enough).

Teeth thickness_mm and RootDiameter_mm I would take a closer look and check if there might be nonlinear effects or interactions I do not have in the model important. 

But again t is glasbowl reading just to look at the actual by predicted plot. I highly recommend to take a look into STIPS e-Learning, especially the courses on "Decision Making With Data" and/or "Correlation and Regression". 

Another short but good resource is Interpreting Regression Results 

Just looking at Actual by Predicted Plot is like just looking at a p-value of 0.055 or a r-square of 0.65. It does not mean  lot without information from other key stats, visualization and some practical/technical knowledge of the underlying system you like to predict or understand.

Additional Tip: In general (about community posts) it is helpful if you describe more in detail what you want to achieve or what you do not understand and want to get an answer for. Your question is quite general and could be answered in many different ways without more information.  Just follow these two guides and it is much easier to answer your questions :

• Getting-fast-helpful-answers-to-your-questions 
• How to deal with an answer - accept solution, kudo, or refine 

[#3 · Accepted solution]

Hi @Sankaramuthu,

Welcome in the Community !

Actual vs. Predicted plots are helpful to visualize two main aspects :

• Visualize model's performances :
  • Bias : Can you spot a specific model pattern, for example if your model is always under or over-predicting values ?
  • Variance/predictive accuracy : What is the relative precision of model's predictions compared to actual values (or what are the residuals magnitude for the response you're trying to predict) ?
• Check unusual patterns and verify model's assumptions for linear regression models :
  • Presence of unusual observations, outliers, leverage points
  • Curvature in the actual vs. predicted plot
  • Statistical significance of the model 

It's best to analyze your model validity and performance using this plot in combination with other residuals plots from the Row Diagnostics section.

I would recommend reading the following ressources to get started :

• About this Effect Leverage plot : Effect Leverage Plots & Statistical Details for Leverage Plots
• About model assumptions verifications : Multiple Regression Residual Analysis and Outliers | Introduction to Statistics | JMP & Regression Model Assumptions | Introduction to Statistics | JMP

You can also learn basics of statistical modeling thanks to the free online course Statistical Thinking (STIPS).

Hope this answer will help you,