cosinor: Cosinor Regression Model for Detecting Seasonality in Yearly...

In agbarnett/season: Seasonal Analysis of Health Data

Cosinor Regression Model for Detecting Seasonality in Yearly Data or Circadian Patterns in Hourly Data

Description

Fits a cosinor model as part of a generalized linear model.

Usage

cosinor(
  formula,
  date,
  data,
  family = gaussian(),
  alpha = 0.05,
  cycles = 1,
  rescheck = FALSE,
  type = "daily",
  offsetmonth = FALSE,
  offsetpop = NULL,
  text = TRUE
)

Arguments

formula	regression formula.
date	a date variable if type=“daily”, or an integer between 1 and 53 if type=“weekly”, or an integer between 1 and 12 if type=“monthly”, or a POSIXct date if type=“hourly”.
data	data set as a data frame.
family	a description of the error distribution and link function to be used in the model. Available link functions: identity, log, logit, cloglog. Note, it must have the parentheses.
alpha	significance level, set to 0.05 (default).
cycles	number of seasonal cycles per year if type=“daily”, “weekly” or “monthly”; number of cycles per 24 hours if type=“hourly”
rescheck	plot the residual checks (TRUE/FALSE), see seasrescheck.
type	“daily” for daily data (default), or “weekly” for weekly data, or “monthly” for monthly data, or “hourly” for hourly data.
offsetmonth	include an offset to account for the uneven number of days in the month (TRUE/FALSE). Should be used for monthly counts (type=“monthly”) (with family=poisson()).
offsetpop	include an offset for the population (optional), this should be a variable in the data frame. Do not log-transform this offset, as the transform is applied by the code.
text	add explanatory text to the returned phase value (TRUE) or return a number (FALSE). Passed to the invyrfraction function.

Details

The cosinor model captures a seasonal pattern using a sinusoid. It is therefore suitable for relatively simple seasonal patterns that are symmetric and stationary. The default is to fit an annual seasonal pattern (cycle=1), but other higher frequencies are possible (e.g., twice per year: cycle=2). The model is fitted using a sine and cosine term that together describe the sinusoid. These parameters are added to a generalized linear model, so the model can be fitted to a range of dependent data (e.g., Normal, Poisson, Binomial). Unlike the nscosinor model, the cosinor model can be applied to unequally spaced data.

Value

Returns an object of class “Cosinor” with the following parts:

call	the original call to the cosinor function.
glm	an object of class glm (see glm).
fitted	fitted values for intercept and cosinor only (ignoring other independent variables).
fitted.plus	standard fitted values, including all other independent variables.
residuals	residuals.
date	name of the date variable (in Date format when type=‘daily’).

Author(s)

Adrian Barnett [e-mail]

References

Barnett, A.G., Dobson, A.J. (2010) Analysing Seasonal Health Data. Springer.

See Also

summary.Cosinor, plot.Cosinor

Examples

## cardiovascular disease data (offset based on number of days in...
## ...the month scaled to an average month length)
data(CVD)
res = cosinor(cvd~1, date='month', data=CVD, type='monthly',
              family=poisson(), offsetmonth=TRUE)
summary(res)
seasrescheck(res$residuals) # check the residuals
## stillbirth data
data(stillbirth)
res = cosinor(stillborn~1, date='dob', data=stillbirth,
              family=binomial(link='cloglog'))
summary(res)
plot(res)
## hourly indoor temperature data
res = cosinor(bedroom~1, date='datetime', type='hourly', data=indoor)
summary(res)
# to get the p-values for the sine and cosine estimates
summary(res$glm)

agbarnett/season documentation built on March 26, 2022, 9:29 a.m.