Skip to contents

Visualization of observations contributing (or not) to estimation

idid_viz_contrib.Rd

Make a graph to visualize observations that can be dropped without changing the point estimate or the standard error of the estimate of interest by more than a given proportion (threshold_change).

Makes a heatmap or a bar chart, depending on the number of dimensions specified.

Usage

idid_viz_contrib(
  reg,
  var_interest,
  var_x,
  var_y,
  contrib_threshold,
  threshold_change = 0.05,
  order = "",
  colors = c("#C25807", "#FBE2C5", "#300D49"),
  keep_labels = TRUE,
  ...
)

Arguments

reg

A regression object.

var_interest

A string. The name of the main variable of interest.

var_x

A variable in the data set used in reg to plot on the x-axis.

var_y

A variable in the data set used in reg to plot on the y-axis (optional). If not specified, produces a bar chart.

contrib_threshold

A numeric (optional). Weight below which observations are deemed to be non-contributing. If not provided, will be determined by running ididvar::idid_contrib_threshold

threshold_change

A double (between 0 and 1). The change threshold in estimate and s.e. when dropping observations.

order

A string (either "x", "y" or "xy") describing whether the graph should be order, along the x or y axis or both. If anything else is specified, no specific ordering will be applied.

colors

A string vector of colors for the palette. I recommend to pass a vector of 3 distinct colors, with a lighter color in the middle, constituting a diverging scale. It allows a clear distinction between contributing and non contributing observations.

keep_labels

A boolean (optional). If FALSE, removes y labels and ticks. This option is useful for panels with a large number of individuals.

...

Additional elements to pass to the regression function when partialling out controls.

Value

A ggplot2 graph of observations contributing to estimation.

If var_y is specified, returns a heatmap whose color describes whether a given observation does not contribute (i.e. can be dropped, along with all lower weight observations, without substantially altering the estimate)

If var_y is not specified, returns a bar chart representing the number of observations that can be dropped in each group.

Examples

reg_ex <- ggplot2::txhousing |>
  lm(formula = log(sales) ~ median + listings + city + as.factor(date))

idid_viz_contrib(reg_ex, "median", year, city, order = "y") +
  ggplot2::labs(x = NULL, y = NULL)
#> Searching for the contribution threshold
#> Warning: Removed 1434 rows containing non-finite outside the scale range
#> (`stat_sum_tile()`).


idid_viz_contrib(reg_ex, "median", var_y = city, order = "y") +
  ggplot2::labs(x = NULL)
#> Searching for the contribution threshold