partition_min_inter decomposes clustered data into individual partitions such as cross-sections and time-series for panel data. It derives an individual intermediate solution for each partition and the pooled data to assess the robustness of the solutions.

partition_min_inter(
  dataset,
  units,
  time,
  cond,
  out,
  n_cut,
  incl_cut,
  intermediate,
  BE_cons,
  WI_cons,
  BE_ncut,
  WI_ncut
)

Arguments

dataset

Calibrated pooled dataset for partitioning and minimization

units

Units defining the within-dimension of data (time series)

time

Periods defining the between-dimension of data (cross sections)

cond

Conditions used for the pooled analysis

out

Outcome used for the pooled analysis

n_cut

Frequency cut-off for designating truth table rows as observed

incl_cut

Inclusion cut-off for designating truth table rows as consistent

intermediate

A vector of directional expectations to derive intermediate solutions

BE_cons

Inclusion (or consistency) thresholds for cross sections. Must be specified as a numeric vector with length equaling the number of cross sections. Numbers correspond to the order of the cross section ID in the data (such as years in ascending order).

WI_cons

Inclusion (or consistency) thresholds for time series. Must be specified as a numeric vector with length equaling the number of time series. Numbers correspond to the order of the time series (unit) ID in the data (such as countries in alphabetical order).

BE_ncut

For cross sections, the minimum number of members needed for declaring a truth table row as relevant as opposed to designating it as a remainder. Must be specified as a numeric vector. Its length should be equal the number of cross sections. The order of thresholds corresponds to the order of the cross sections in the data defined by the cross-section ID in the dataset (such as years in ascending order).

WI_ncut

For time series, the minimum number of members needed for declaring a truth table row as relevant as opposed to designating it as a remainder. Must be specified as a numeric vector. Its length should be equal the number of time series. The order of thresholds corresponds to the order of the of the time-series (unit) ID in the dataset (such as countries in alphabetical order).

Value

A dataframe summarizing the partition-specific and pooled solutions with the following columns:

  • type: The type of the partition. pooled are rows with information on the pooled data; between is for cross-section partitions; within is for time-series partitions.

  • partition: Specific dimension of the partition at hand. For between-dimension, the unit identifiers are included here (argument units). For the within-dimension, the time identifier are listed (argument time). The entry is - for the pooled data without partitions.

  • solution: The solution derived for the partition or the pooled data. Absence of a condition is denoted by the ~ sign.

  • model: Running ID for models. In the presence of model ambiguity, each model has its own row with its individual solution and parameters. The rest of the information in the row is duplicated, for example by having two rows for the within-partition 1996. The column model highlights the presence of model ambiguity by numbering all models belonging to the same solution. For example, if three consecutive rows are numbered 1, 2 and 3, then these rows belong to the same solution and represent model ambiguity. If a 1 in a row is followed by another 1, then there is no model ambiguity.

  • consistency: The consistency score (a.k.a. inclusion score) for the partition of the data or the pooled data.

  • coverage: The coverage score for the partition of the data or the pooled data.

Examples

data(Schwarz2016) Schwarz_inter <- partition_min_inter( Schwarz2016, units = "country", time = "year", cond = c("poltrans", "ecotrans", "reform", "conflict", "attention"), out = "enlarge", n_cut = 1, incl_cut = 0.8, intermediate = c("1", "1", "1", "1", "1"))