This function takes an object returned from the svyglm function and generates a clean summary table for a statistical report. Similar to tabglm, but for survey data. Relies heavily on the 'survey' package [1,2].
tabglm.svy(svyglmfit, latex = FALSE, xlabels = NULL, ci.beta = TRUE, inference = "wald.t",
decimals = 2, p.decimals = c(2, 3), p.cuts = 0.01, p.lowerbound = 0.001,
p.leading0 = TRUE, p.avoid1 = FALSE, basic.form = FALSE, intercept = TRUE,
n = FALSE, events = FALSE, greek.beta = FALSE, binary.compress = TRUE,
bold.colnames = TRUE, bold.varnames = FALSE, bold.varlevels = FALSE,
predictor.colname = "Variable")An object returned from svyglm function call [1,2].
If TRUE, object returned is formatted for printing in LaTeX using xtable [3]; if FALSE, formatted for copy-and-pasting from RStudio into a word processor.
Optional character vector to label the x variables and their levels. If unspecified, the function uses the variable names and values themselves.
If TRUE, the table returned will include a column for Wald 95% confidence interval for the regression coefficients.
If "wald.t", confidence intervals and p-values are based on t distributions with degrees of freedom given by svyglmfit$df.residual; if "wald.z", confidence intervals and p-values are based on z distributions.
Number of decimal places for numeric values in the table (except p-values).
Number of decimal places for p-values. If a vector is provided rather than a single value, number of decimal places will depend on what range the p-value lies in. See p.cuts.
Cut-point(s) to control number of decimal places used for p-values. For example, by default p.cuts is 0.1 and p.decimals is c(2, 3). This means that p-values in the range [0.1, 1] will be printed to two decimal places, while p-values in the range [0, 0.1) will be printed to three decimal places.
Controls cut-point at which p-values are no longer printed as their value, but rather <lowerbound. For example, by default p.lowerbound is 0.001. Under this setting, p-values less than 0.001 are printed as <0.001.
If TRUE, p-values are printed with 0 before decimal place; if FALSE, the leading 0 is omitted.
If TRUE, p-values rounded to 1 are not printed as 1, but as >0.99 (or similarly depending on values for p.decimals and p.cuts).
If TRUE, there is no attempt to neatly format factor variables and their levels, and the table returned is very similar to what you see when you run summary(glmfit).
If FALSE, the table returned will not include a row for the intercept.
If TRUE, the table returned will include a column for sample size.
If TRUE, the table returned will include a column for number of events observed. Only meaningful when the outcome variable is binary.
If TRUE, column headings refer to regression parameters as Greek letter beta rather than Beta. Only used when latex input is set to TRUE.
If TRUE, only one row of the table is dedicated to parameter estimates for each binary factor predictor. If FALSE, the table displays separate rows for the variable name and the two levels for each binary factor predictor, much like the presentation for factor variables with more than two levels.
If TRUE, column headings are printed in bold font. Only applies if latex = TRUE.
If TRUE, variable names in the first column of the table are printed in bold font. Only applies if latex = TRUE.
If TRUE, levels of each factor variable are printed in bold font. Only applies if latex = TRUE and there is at least one factor variable included as a predictor.
Character string with desired column heading for the column of predictors.
A character matrix that summarizes the fitted generalized linear model. If you click on the matrix name under "Data" in the RStudio Workspace tab, you will see a clean table that you can copy and paste into a statistical report or manuscript. If latex is set to TRUE, the character matrix will be formatted for inserting into an Sweave or Knitr report using the xtable package [3].
The function should work well with categorical predictors (factors), provided they are not ordered. For ordered factors, just convert to unordered before creating the svyglm object to pass to tabglm.svy. Note that you can define the levels of an unordered factor to control, which dictates which level is used as the reference group in regression models. For example, suppose a factor variable x takes values "low", "medium", and "high". If you write x = factor(x = x, levels = c("low", "medium", "high")), then you can run levels(x) to see that the levels are now arranged "low", "medium", "high". It is still a regular factor, but now if you use x as a predictor in a call to svyglm, "low" will be the reference group.
Interaction terms are compatible with tabglm.svy, but the table will be formatted a little differently if interaction terms are present. Basically including an interaction is equivalent to setting basic.form to TRUE. All variable names and levels will be exactly as they appear when you run summary(svyglmfit), where svyglmfit is the object returned from a call to svyglm.
1. Lumley T (2012). survey: analysis of complex survey samples. R package version 3.28-2, https://cran.r-project.org/package=survey.
2. Lumley T (2014). Analysis of complex survey samples. Journal of Statistical Software 9(1): 1-19.
3. Dahl DB (2013). xtable: Export tables to LaTeX or HTML. R package version 1.7-1, https://cran.r-project.org/package=xtable.
Acknowledgment: This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE-0940903.
svydesign
svyglm
tabfreq,
tabmeans,
tabmedians,
tabmulti,
tabglm,
tabcox,
tabgee,
tabfreq.svy,
tabmeans.svy,
tabmedians.svy,
tabmulti.svy