Three parameters of the normal linear model
A normal linear model,
μy = β0 + β1x
σy = σ
involves 3 parameters, β0, β1 and σ. These parameters provide considerable flexibility to the model.
Drag the three red arrows to adjust the parameters of the normal linear model.
Click Take sample a few times to verify that approximately 95% of values are within the grey band.
(Note that the values of X are not fixed in this example — they vary from sample to sample. The normal linear model does not attempt to describe variability in X, though a standard univariate distribution such as a normal distribution might fit the distribution of X in this example.)
Interpreting the model's slope and intercept
The most important parameters of a linear model are its slope, β1, and intercept, β0. These can be interpreted in a similar way to the slope and intercept of the least squares lines that were fitted to data in an earlier chapter.
Context | Interpretation of β1 | Interpretation of β0 |
---|---|---|
Y = Yield of wheat per acre X = Fertiliser (kg per m2) |
Increase in mean yield per acre for each additional kg/m2 of fertiliser | Mean yield per acre if no fertiliser is used |
Y = Exam mark X = Hours of study by student before exam |
Increase in expected mark for each additional hour of study | Expected mark if there is no study |
Y = Hospital stay (days) X = Age of patient |
Average extra days in hospital per extra year of age | Average days in hospital at age 0. Not particularly meaningful here. |
Esimating the parameters by eye
The regression line (i.e. the straight line showing how the mean depends on x) and the band that is 2σ above and below it are a good way to understand the normal linear model. Indeed they can be used as an informal way to estimate the parameters of the model 'by eye'. (We will give better methods in the next section.)
Artificial data
A normal linear model might be used to describe how the response depends on the explanatory variable.
In the next section, we will explain how to objectively estimate the parameters to match a data set. Click Best values to see these 'best' parameter values.
If there are fewer values or if the relationship is weaker, it is harder to position the band by eye.
Body and heart weights of cats
The data set below shows the body and heart weights of 47 female cats. The data might be used to predict heart weight from body weight since the latter can be easily measured from live cats!
A normal linear model might be used to describe how heart weight depends on body weight.
In the next section, we will explain how to objectively estimate the parameters to match a data set.