create_ITSA: Estimate an effect of intervention on every Viva Insights metric in input file by applying single-group Interrupted Time-Series Analysis (ITSA)

Description

r lifecycle::badge('experimental')

This function implements ITSA method described in the paper 'Conducting interrupted time-series analysis for single- and multiple-group comparisons', Ariel Linden, The Stata Journal (2015), 15, Number 2, pp. 480-500

This function further requires the installation of 'sandwich', 'portes', and 'lmtest' in order to work. These packages can be installed from CRAN using install.packages().

Usage

create_ITSA(
  data,
  before_start = min(as.Date(data$Date, "%m/%d/%Y")),
  before_end,
  after_start,
  after_end = max(as.Date(data$Date, "%m/%d/%Y")),
  ac_lags_max = 7,
  return = "table"
)

Arguments

data

Person Query as a dataframe including date column named Date. This function assumes the data format is MM/DD/YYYY as is standard in a Viva Insights query output.

before_start

Start date of 'before' time period in MM/DD/YYYY format as character type. Before time period is the period before the intervention (e.g. training program, re-org, shift to remote work) occurs and bounded by before_start and before_end parameters. Longer period increases likelihood of achieving more statistically significant results. Defaults to earliest date in dataset.

before_end

End date of 'before' time period in MM/DD/YYYY format as character type.

after_start

Start date of 'after' time period in MM/DD/YYYY format as character type. After time period is the period after the intervention occurs and bounded by after_start and after_end parameters. Longer period increases likelihood of achieving more statistically significant results. Defaults to date after before_end.

after_end

End date of 'after' time period in MM/DD/YYYY format as character type. Defaults to latest date in dataset.

ac_lags_max

maximum lag for autocorrelation test. Default is 7

return

String specifying what output to return. Defaults to "table". Valid return options include:

'plot': return a list of plots.
'table': return data.frame with estimated models' coefficients and their corresponding p-values You should look for significant p-values in beta_2 to indicate an immediate treatment effect, and/or in beta_3 to indicate a treatment effect over time

Author

Aleksey Ashikhmin alashi@microsoft.com

Details

This function uses the additional package dependencies 'sandwich' and 'lmtest'. Please install these separately from CRAN prior to running the function.

As of May 2022, the 'portes' package was archived from CRAN. The dependency has since been removed and dependent functions Ljungbox() incorporated into the wpa package.

Examples

Run this code

# Returns summary table

create_ITSA(
  data = sq_data,
  before_start = "11/03/2019",
  before_end = "12/15/2019",
  after_start = "12/29/2019",
  after_end = "1/26/2020",
  ac_lags_max = 7,
  return = "table")

# Returns list of plots

plot_list <-
  create_ITSA(
    data = sq_data,
    before_start = "11/03/2019",
    before_end = "12/15/2019",
    after_start = "12/29/2019",
    after_end = "1/26/2020",
    ac_lags_max = 7,
    return = 'plot')

# Extract a plot as an example
plot_list$Workweek_span