Date: 2019-10-07
This repository contains simple functions in a package format, and is in active development. This project is commissioned by the UN Population Division and financed by the Bill and Melinda Gates Foundation as part of the Making Family Planning Count project. Work is also done in collaboration with Sean Fennell, and Jose Manuel Aburto, Ilya Kashnitsky, Marius Pascariu with minor contributions from several more (thank you!). This work is licensed under the Creative Commons Attribution-ShareAlike 3.0 IGO (CC BY-SA 3.0 IGO).
If you detect a bug or have a suggestion please notify us using the Issues tab on github. Even better if you fix it and make a pull request! See CONTRIBUTING.md for more tips on reporting bugs or offering patches.
You can load the DemoTools package in R like so:
# install.packages("devtools")
library(devtools)
install_github("timriffe/DemoTools")(if either of the first two icons at the top of this README are red, then this might not be working at the moment. You can assume we're fixing it. If they're green, then it'll probably work.)
Sometime soon there will be an overhaul of function names. We plan to switch to snake case, with method families as the first element. This is to make naming more regular and memorable, and also to activate autocomplete in RStudio or similar.
We'll soon add a primer here, but for now you can get started by loading the package and calling up help files, which contain worknig examples to demonstrate usage and options.
library(DemoTools)
# interesting top-level functions include:
# for age heaping:
?Whipple
?Myers
?Bachi
?CoaleLi
?Noumbissi
?Spoorenberg
?KannistoHeap #(Kannisto's old-age heaping index)
?Jdanov (Jdanov's old-age heaping index)
?heapify # induce heaping, to test evaluation functions
# test if 5-year smoothing recommended:
?zero_pref_sawtooth # is heaping much worse on 0s than on 5s?
?five_year_roughness # measure of total roughness
# other age-structure quality measures:
?ageRatioScore # methods including "UN", "Zelnick", "Ramachandran"
?sexRatioScore
?ageSexAccuracy # methods including "UN", "Zelnick", "Ramachandran", and "Das Gupta"
# Comparison methods
?IRD (index of relative difference)
?ID (index of dissimilarity)
?survRatioError
# graduation methods
?sprague
?beers # methods including "ord" and "mod", as well as johnson option for young ages
?grabill
?splitMono
?monoCloseout
?splitOscillate # accepting e.g. spragueSimple, beersSimple as split methods
# various smoothing methods
# * for 5-year age groups
?agesmth # including Carrier-Farrag, Arriaga, Karup-King-Newton, United Nations, Strong, Zigzag, and MAV methods
# * for single ages
?agesmth1 # including loess and polynomial
?spencer
?zelnik
# various lifetable evaluation and calculation functions
?ADM # and ?RDM, implementing PAS LIFIT
?LTabr # with fine control over a(x) assumptions, extrapolation, and open age groups
# interpolation
?interp (arithmetic, logarithmic, power)
# redistribution
?OPAG_simple # increase population open age, redistributing using a supplied standard.
?rescaleAgeGroups (including for cases of different age groupings)
These top-level functions have implied an even larger set of simple utilities, which itself is growing fast. Presently top-level + utilities = 118 documented functions, with more in development.
Presently all functions are in a testing phase, but the aim is to end up with a set of robust generic functions around which wrappers can be easily built for various institutional data production needs. As-is, these functions may also be useful for DIY demographers. This set of methods is a cherry-pick from legacy methods collections, including PAS, DAPPS, MPCDA, MortPack, IREDA, UN Manual X, G. Feeney Spreadsheets, formulas found in Siegel and Swanson or Shyrock and Siegel, and various (apparent) first-implementations from formulas in papers, or ad hoc DIY approximations from old pros.
Every time this repository is updated the entire code base is rebuilt on a server somewhere, and undergoes a series of checks. This happens on a Linux machine and on a Windows machine. Any warnings or errors in these builds will yield a red fail tag, and successes are green passes. Code coverage indicates what percentage of lines of code undergo formal unit testing of some kind.