Introduction to mpindex

In mpindex: Multidimensional Poverty Index (MPI)

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

This guide presents an illustration of a simple application of mpindex package for computing Multidimensional Poverty Index (MPI) using the Alkire-Foster (AF) counting method developed by Sabina Alkire and James Foster.

1. Installation

To install the mpindex package from CRAN:

install.packages('mpindex')

If you want to get the latest development version of mpindex, install it from GitHub. Note that you may need to install devtools.

# install.packages("devtools")
devtools::install_github('yng-me/mpindex')

Load the package once you have successfully completed the installation.

library(mpindex)

2. MPI specifications

The initial step is to prepare an MPI specification file which will serve as references in the computation as well as generation of output in the later part of the process. This file should contain information about MPI dimensions, indicators and their corresponding weights.

This file should also be easy to create using the most common and accessible file types such as .xlsx (Excel), .json, .csv, or .txt (TSV).

Built-in specification files

For convenience, mpindex has included built-in specification files (in different formats). Each file contains dimensions, indicators, weight, and other relevant information of the Global MPI.

To see the list of files available:

system.file("extdata", package = "mpindex") |> list.files()

To use a built-in specification file, say the .csv file, use below script to first get the full path of the file.

specs_file <- system.file("extdata", "global-mpi-specs.csv", package = "mpindex")

read.csv(specs_file) |> 
  gt::gt() |> 
  gt::tab_header(
    title = 'Global MPI – Dimensions, Indicators, Deprivation Cutoffs, and Weights'
  ) |> 
  gt::tab_options(
    table.width = '100%',
    table.font.size = 12,
  ) |> 
  gt::tab_footnote('Source: Alkire, S., Kanagaratnam, U. and Suppa, N. (2020). ‘The global Multidimensional Poverty Index (MPI): 2020 revision’, OPHI MPI Methodological Note 49, Oxford Poverty and Human Development Initiative, University of Oxford.') |> 
  gt::fmt_number(
    columns = 4,
    decimals = 3
  )

User-defined specification file

You can also define and create your own specification file if you prefer to or if you happen to use different sets of dimensions and indicators. At the minimum, however, this file should contain the following columns/variables:

• Dimension,
• Indicator,
• Variable,
• Weight,
• Description (optional).

Note that the order in which you put these columns does not matter and also the names are not case sensitive, but make sure to spell the column names correctly.

You may download the template here if you do not want to start from scratch: MPI specification file sample.

Using define_mpi_specs

Once you have prepared the specification file, load it using the define_mpi_specs function (type ?define_mpi_specs for a help text).

For demonstration purposes, we will use the built-in specification file, discussed above.

specs_file <- system.file("extdata", "global-mpi-specs.csv", package = "mpindex")
define_mpi_specs(specs_file)

define_mpi_specs requires a specification file as its first argument. The default poverty cutoff is set to 1/3 (based on Global MPI). You can also define a list of poverty cutoffs by specifying in the .poverty_cutoffs argument to achieve gradient MPIs.

define_mpi_specs(
  .mpi_specs_file = specs_file, 
  .poverty_cutoffs = c(1/3, 0.2, 0.8)
)

If your dataset contains unique ID, like uuid, it is recommended to define it here using the .uid argument.

define_mpi_specs(
  .mpi_specs_file = specs_file, 
  .uid = 'uuid'
)

You can also set the aggregation level. Make sure it corresponds to the column name present in your dataset (see ?df_househod and more below).

define_mpi_specs(
  .mpi_specs_file = specs_file, 
  .poverty_cutoffs = c(1/3, 0.2, 0.8),
  .uid = 'uuid',
  .aggregation = 'class'
)

.unit_of_analysis, .source_of_data, and .names_separator are merely used for auto labels when generating the output later.

Note: define_mpi_specs returns a data frame of MPI specs defined in the specs file. By default, it saves a global option named mpi_specs which can be accessed using getOption('mpi_specs').

For our demonstration, we will use the use_global_mpi_specs() wrapper function, which yields the same result as calling define_mpi_specs and passing the built-in specs file.

use_global_mpi_specs(
  .uid = 'uuid',
  .aggregation = 'class'
)

use_global_mpi_specs(
  .uid = 'uuid',
  .aggregation = 'class'
)

3. Data preparation

The user of mpindex is assumed to have basic familiarity with the concept of tidy data as well as able to perform data wrangling and transformation using the tidyverse ecosystem. Under the hood, mpindex uses dplyr verbs to perform data manipulation.

We also assume that your dataset is already tidy and ready for analysis. See R for Data Science by Hadley Wickham and Garrett Grolemund if you need a refresher.

Dataset

For this demonstration, we will use two (2) synthetic datasets available within the package:

• df_household household-level data (type ?df_household for more info)
• df_household_roster individual-level data (type ?df_household_roster for more info)

library(dplyr)

glimpse(df_household)

glimpse(df_household_roster)

4. Creating deprivation profile

Using define_deprivation

First, we need to create an empty list, and name it deprivation_profile (but feel free to name it whatever you like).

deprivation_profile <- list()

To create a deprivation profile for each indicator, we use the define_deprivation function (see ?define_deprivation) and add to the deprivation_profile list we created above. Make sure that the deprivation profile for each indicator matches the variable name declared in the specification file.

1. Heath dimension

1.1. Nutrition

For this indicator, we use the df_household_roster dataset. By default, define_deprivation sets the .collapse = FALSE. Since we need to collapse it to the household level, we need to set .collapse = TRUE.

deprivation_profile$nutrition <- df_household_roster |> 
  define_deprivation(
    .indicator = nutrition,
    .cutoff = undernourished == 1 & age < 70,
    .collapse = TRUE
  )

1.2. Child mortality

For child mortality, we use the df_household dataset. But unlike in nutrition, we do not need to provide the .collapse argument since it is not applicable here.

deprivation_profile$child_mortality <- df_household |> 
  define_deprivation(
    .indicator = child_mortality,
    .cutoff = with_child_died == 1
  )

2. Education dimension

2.1. Years of schooling

deprivation_profile$year_schooling <- df_household_roster |> 
  define_deprivation(
    .indicator = year_schooling,
    .cutoff = completed_6yrs_schooling == 2,
    .collapse = TRUE
  )

2.2. School attendance

deprivation_profile$school_attendance <- df_household_roster |> 
  define_deprivation(
    .indicator = school_attendance,
    .cutoff = attending_school == 2 & age %in% c(5:24),
    .collapse = TRUE
  )

3. Living standards dimension

3.1. Cooking fuel

deprivation_profile$cooking_fuel <- df_household |> 
  define_deprivation(
    .indicator = cooking_fuel,
    .cutoff = cooking_fuel %in% c(4:6, 9)
  )

3.2. Sanitation

deprivation_profile$sanitation <- df_household |> 
  define_deprivation(
    .indicator = sanitation,
    .cutoff = toilet > 1
  )

3.3. Drinking water

deprivation_profile$drinking_water <- df_household |> 
  define_deprivation(
    .indicator = drinking_water,
    .cutoff = drinking_water == 2
  )

3.4. Electricity

deprivation_profile$electricity <- df_household |> 
  define_deprivation(
    .indicator = electricity,
    .cutoff = electricity == 2
  )

3.5. Housing

deprivation_profile$housing <- df_household |> 
  define_deprivation(
    .indicator = housing,
    .cutoff = roof %in% c(5, 7, 9) | walls %in% c(5, 8, 9, 99) == 2 | floor %in% c(5, 6, 9)
  )

3.6. Assets

For this indicator, we need additional transformation.

deprivation_profile$assets <- df_household |> 
  mutate_at(vars(starts_with('asset_')), ~ if_else(. > 0, 1L, 0L)) |> 
  mutate(
    asset_phone = if_else(
      (asset_telephone + asset_mobile_phone) > 0, 
      1L, 
      0L
    )
  ) |> 
  mutate(
    with_hh_conveniences = (
      asset_tv + asset_phone + asset_computer + 
        asset_animal_cart + asset_bicycle + 
        asset_motorcycle + asset_refrigerator) > 1,
    with_mobility_assets = (asset_car + asset_truck) > 0
  ) |> 
  define_deprivation(
    .indicator = assets,
    .cutoff = !(with_hh_conveniences & with_mobility_assets)
  )

5. Computing the MPI

Using compute_mpi

After completing the deprivation profile, use the compute_mpi function and pass the deprivation_profile list as the first argument.

mpi_result <- df_household |>
  compute_mpi(deprivation_profile)

names(mpi_result)

Outputs

1. The MPI

mpi_result$index

mpi_result$index |>
  rename(Class = 1) |> 
  gt::gt() |> 
  gt::tab_header(
    title = 'MPI Results using 33% Poverty Cutoff'
  ) |> 
  gt::fmt_number(
    columns = 3:5,
    decimals = 3
  ) |> 
  gt::tab_options(
    table.width = '100%',
    table.font.size = 12,
  )

2. Contribution by dimension

mpi_result$contribution

gtx <- function(.gt, .decimals = 1, .offset = 0) {
  d01_cp <- 3:4 + .offset
  d02_cp <- 5:6 + .offset
  d03_cp <- 7:12 + .offset

  .gt |> 
    gt::tab_spanner(
      label = "Health",
      columns = d01_cp
    ) |> 
    gt::tab_spanner(
      label = "Education",
      columns = d02_cp
    ) |> 
    gt::tab_spanner(
      label = "Living Standards",
      columns = d03_cp
    ) |> 
    gt::fmt_number(
      columns = c(d01_cp, d02_cp, d03_cp),
      decimals = .decimals
    ) |> 
    gt::tab_options(
      table.font.size = 12,
    )
}

mpi_result$contribution |> 
  gt::gt() |> 
  gt::tab_header(
    title = 'Contribution by Dimenstion and Indicator to MPI using 33% Poverty Cutoff'
  ) |> 
  gtx()

3. Headcount ratio

mpi_result$headcount_ratio$uncensored

mpi_result$headcount_ratio$uncensored |> 
  ungroup() |> 
  gt::gt() |> 
  gt::tab_header(
    title = 'Uncensored Headcount Ratio'
  ) |> 
  gtx(.decimals = 3)

mpi_result$headcount_ratio$censored

mpi_result$headcount_ratio$censored |> 
  ungroup() |> 
  gt::gt() |> 
  gt::tab_header(
    title = 'Censored Headcount Ratio using 33% Poverty Cutoff'
  ) |> 
  gtx(.decimals = 3)

4. Deprivation matrix (first 6 observations)

mpi_result$deprivation_matrix$uncensored |> head()

mpi_result$deprivation_matrix$uncensored |> 
  ungroup() |> 
  head() |> 
  rename_all(~ stringr::str_remove(., '^(Health|Education|Living Standards)>')) |> 
  gt::gt() |> 
  gt::tab_header(
    title = 'Uncensored Deprivation Matrix '
  ) |> 
  gtx(.decimals = 0, .offset = 1) |> 
  gt::fmt_number(
    columns = 3,
    decimals = 3
  ) 

mpi_result$deprivation_matrix$censored |> head()

mpi_result$deprivation_matrix$censored |> 
  ungroup() |> 
  head() |> 
  rename_all(~ stringr::str_remove(., '^(Health|Education|Living Standards)>')) |> 
  gt::gt() |> 
  gt::tab_header(
    title = 'Censored Deprivation Matrix using 33% Poverty Cutoff'
  ) |> 
  gtx(.decimals = 0, .offset = 1) |> 
  gt::fmt_number(
    columns = 3,
    decimals = 3
  )

Note: Deprivation matrices are included by default when you run compute_mpi. If you want to exclude these in your output, set .include_deprivation_matrix equal to FALSE.

6. Saving output

You may also save your output into an Excel file. You may choose to format the output or retain its tidy format by setting the formatted_output parameter appropriately.

# Formatted output
save_mpi(mpi_result, .filename = 'MPI Sample Output')

# Not formatted
save_mpi(mpi_result, .filename = 'MPI Sample Output (no format)', .formatted_output = FALSE)

Full script

# ----------------------------------
# Load MPI specs from the built-in specs file 
use_global_mpi_specs(
  .uid = 'uuid',
  .aggregation = 'class'
)

# ----------------------------------
# Create an empty list to store deprivation profile for each indicator
deprivation_profile <- list()

deprivation_profile$nutrition <- df_household_roster |>
  define_deprivation(
   .indicator = nutrition,
   .cutoff = undernourished == 1 & age < 70,
   .collapse = TRUE
  )

deprivation_profile$child_mortality <- df_household |>
  define_deprivation(
   .indicator = child_mortality,
   .cutoff = with_child_died == 1
  )

deprivation_profile$year_schooling <- df_household_roster |>
  define_deprivation(
   .indicator = year_schooling,
   .cutoff = completed_6yrs_schooling == 2,
   .collapse = TRUE
  )

deprivation_profile$school_attendance <- df_household_roster |>
  define_deprivation(
   .indicator = school_attendance,
   .cutoff = attending_school == 2 & age %in% c(5:24),
   .collapse = TRUE
  )

deprivation_profile$cooking_fuel <- df_household |>
  define_deprivation(
   .indicator = cooking_fuel,
   .cutoff = cooking_fuel %in% c(4:6, 9)
  )

deprivation_profile$sanitation <- df_household |>
  define_deprivation(
   .indicator = sanitation,
   .cutoff = toilet > 1
  )

deprivation_profile$drinking_water <- df_household |>
  define_deprivation(
   .indicator = drinking_water,
   .cutoff = drinking_water == 2
  )

deprivation_profile$electricity <- df_household |>
  define_deprivation(
   .indicator = electricity,
   .cutoff = electricity == 2
  )

deprivation_profile$housing <- df_household |>
  define_deprivation(
   .indicator = housing,
   .cutoff = roof %in% c(5, 7, 9) | 
     walls %in% c(5, 8, 9, 99) == 2 | 
     floor %in% c(5, 6, 9)
  )

deprivation_profile$assets <- df_household |>
  dplyr::mutate_at(
    dplyr::vars(dplyr::starts_with('asset_')), 
    ~ dplyr::if_else(. > 0, 1L, 0L)
  ) |>
  dplyr::mutate(
   asset_phone = dplyr::if_else(
     (asset_telephone + asset_mobile_phone) > 0,
     1L,
     0L
   )
  ) |>
  dplyr::mutate(
   with_hh_conveniences = (
     asset_tv + asset_phone + asset_computer +
       asset_animal_cart + asset_bicycle +
       asset_motorcycle + asset_refrigerator) > 1,
   with_mobility_assets = (asset_car + asset_truck) > 0
  ) |>
  define_deprivation(
   .indicator = assets,
   .cutoff = !(with_hh_conveniences & with_mobility_assets)
  )

# ----------------------------------
# Compute the MPI
mpi_result <- df_household |>
  compute_mpi(deprivation_profile)

# ----------------------------------
# You may also save your output into an Excel file

# Formatted output
# save_mpi(mpi_result, .filename = 'MPI Sample Output', .include_specs = T)

# Not formatted
save_mpi(
  mpi_result, 
  .filename = 'MPI Sample Output (no format)', 
  .formatted_output = FALSE, 
  .include_specs = TRUE
)

mpindex documentation built on May 29, 2024, 6:54 a.m.