Fit a polynomial surface determined by one or more numerical predictors, using local fitting.

```
loess(formula, data, weights, subset, na.action, model = FALSE,
span = 0.75, enp.target, degree = 2,
parametric = FALSE, drop.square = FALSE, normalize = TRUE,
family = c("gaussian", "symmetric"),
method = c("loess", "model.frame"),
control = loess.control(…), …)
```

formula

a formula specifying the numeric response and one to four numeric predictors (best specified via an interaction, but can also be specified additively). Will be coerced to a formula if necessary.

data

an optional data frame, list or environment (or object
coercible by `as.data.frame`

to a data frame) containing
the variables in the model. If not found in `data`

, the
variables are taken from `environment(formula)`

,
typically the environment from which `loess`

is called.

weights

optional weights for each case.

subset

an optional specification of a subset of the data to be used.

na.action

the action to be taken with missing values in the
response or predictors. The default is given by
`getOption("na.action")`

.

model

should the model frame be returned?

span

the parameter \(\alpha\) which controls the degree of smoothing.

enp.target

an alternative way to specify `span`

, as the
approximate equivalent number of parameters to be used.

degree

the degree of the polynomials to be used, normally 1 or 2. (Degree 0 is also allowed, but see the ‘Note’.)

parametric

should any terms be fitted globally rather than locally? Terms can be specified by name, number or as a logical vector of the same length as the number of predictors.

drop.square

for fits with more than one predictor and
`degree = 2`

, should the quadratic term be dropped for particular
predictors? Terms are specified in the same way as for
`parametric`

.

normalize

should the predictors be normalized to a common scale if there is more than one? The normalization used is to set the 10% trimmed standard deviation to one. Set to false for spatial coordinate predictors and others known to be on a common scale.

family

if `"gaussian"`

fitting is by least-squares, and if
`"symmetric"`

a re-descending M estimator is used with Tukey's
biweight function. Can be abbreviated.

method

fit the model or just extract the model frame. Can be abbreviated.

control

control parameters: see `loess.control`

.

…

control parameters can also be supplied directly
(*if* `control`

is not specified).

An object of class `"loess"`

.

Fitting is done locally. That is, for the fit at point \(x\), the
fit is made using points in a neighbourhood of \(x\), weighted by
their distance from \(x\) (with differences in ‘parametric’
variables being ignored when computing the distance). The size of the
neighbourhood is controlled by \(\alpha\) (set by `span`

or
`enp.target`

). For \(\alpha < 1\), the
neighbourhood includes proportion \(\alpha\) of the points,
and these have tricubic weighting (proportional to \((1 -
\mathrm{(dist/maxdist)}^3)^3\)). For
\(\alpha > 1\), all points are used, with the
‘maximum distance’ assumed to be \(\alpha^{1/p}\)
times the actual maximum distance for \(p\) explanatory variables.

For the default family, fitting is by (weighted) least squares. For
`family="symmetric"`

a few iterations of an M-estimation
procedure with Tukey's biweight are used. Be aware that as the initial
value is the least-squares fit, this need not be a very resistant fit.

It can be important to tune the control list to achieve acceptable
speed. See `loess.control`

for details.

W. S. Cleveland, E. Grosse and W. M. Shyu (1992) Local regression
models. Chapter 8 of *Statistical Models in S* eds J.M. Chambers
and T.J. Hastie, Wadsworth & Brooks/Cole.

`lowess`

, the ancestor of `loess`

(with
different defaults!).

# NOT RUN { cars.lo <- loess(dist ~ speed, cars) predict(cars.lo, data.frame(speed = seq(5, 30, 1)), se = TRUE) # to allow extrapolation cars.lo2 <- loess(dist ~ speed, cars, control = loess.control(surface = "direct")) predict(cars.lo2, data.frame(speed = seq(5, 30, 1)), se = TRUE) # }