# Use DFO data in your code This page is for people who analyse data by writing code in Python or R. It shows how to take a file downloaded from the DFO, open it in your code, and work with the numbers. The example uses aid for health in Nigeria. It keeps only official development assistance, totals it by year, and saves the result together with the source's citation details so your work stays traceable. If you don't write code, you can open the CSV download in Excel instead, or use DFO Studio to make charts without any code. ## Example output A download from [`/data`](/data) is a ZIP file with a README, the data file, a table of code names, `SOURCES.md`, `CITATION.cff`, and `datapackage.json`. Load it into pandas or R, filter it to one flow type, and keep its citation with your results. For the Nigeria health download, the pandas script prints official development assistance (ODA) gross disbursements in millions of current US dollars. ```text time_period 2020 592.4 2021 753.4 2022 1115.5 2023 768.9 2024 843.9 Name: value, dtype: float64 ``` ## Before you start - Python with `pandas` and `pyarrow`, or R with `arrow`, `readr`, and `dplyr`. - A download. On `/data`, search for "CRS disbursements to Nigeria for health". Open **Download options**, the arrow beside the download button, choose **Parquet** under **Format**, and leave **README** and **Sources & Citations** ticked. Select **Download Parquet**. [Get data](../get-started/get-data.md) walks through the search. The file name carries the sources, the date the DFO first prepared the version, the format, and the first six characters of the `content_version`, the id of the exact rows in the file. Your numbers match this page when your file is `dfo_oecd-creditor-reporting-system__parquet_b785be.zip`. A different last part means the rows changed, whether the source was revised or the DFO's processing changed. [The data bundle](../data/bundle.md#file-names) explains each part. ## Steps 1. Unzip the download into a folder. ```console $ unzip dfo_*_parquet_b785be.zip -d nga-health extracting: nga-health/data/oecd-creditor-reporting-system.parquet extracting: nga-health/data/codes.csv extracting: nga-health/README.md extracting: nga-health/SOURCES.md extracting: nga-health/CITATION.cff extracting: nga-health/datapackage.json ``` 1. Load the data, then sum ODA by year. A search can return several flow types, and summing them mixes aid with other finance, so the script keeps only ODA rows before summing. Run the script from the folder containing `nga-health`. The script prints the output above. ```python import shutil from pathlib import Path import pandas as pd bundle = Path("nga-health") df = pd.read_parquet(bundle / "data" / "oecd-creditor-reporting-system.parquet") oda = df[df["crsTypeOfFlow"] == "CRSTypeOfFlow10"] by_year = oda.groupby("time_period")["value"].sum().div(1e6).round(1) print(by_year.tail(5)) by_year.to_csv("nga-health-oda.csv") shutil.copy(bundle / "CITATION.cff", "nga-health-oda.CITATION.cff") ``` Each row in this CRS download represents one donor, one year, and one series. `value` uses the currency named in `unit`, here `CurrencyUSD`. The `crs…` columns (`crsPurpose`, `crsCooperationModality`, `crsTypeOfFinance`, and `crsTypeOfFlow`) hold the codes for the series' purpose, modality, type of finance, and type of flow, and each has a `_label` column beside it with the code's name. `CRSTypeOfFlow10` is official development assistance. `data/codes.csv` gives the plain name of each code, such as `country/KEN` for Kenya, and `SOURCES.md` gives each source's licence and the steps the DFO applied. 1. Keep the citation with your results. The script in step 2 copies `CITATION.cff` next to `nga-health-oda.csv`. To cite the data in a report, use a snapshot link in the form [Cite and licence](../about/cite-and-licence.md) indicates. ### Load the CSV instead Choose **CSV** under **Format** and select **Download CSV**. The zip name ends `_csv_b785be.zip`, and it holds `data/oecd-creditor-reporting-system.csv` in place of the Parquet file. Unzip it with `-d nga-health-csv`. ```python import pandas as pd df = pd.read_csv("nga-health-csv/data/oecd-creditor-reporting-system.csv", dtype={"time_period": str}) print(df[["financingProvider", "time_period", "value", "unit"]].head(3)) ``` ```text financingProvider time_period value unit 0 country/DEU 2015 41720.0 CurrencyUSD 1 country/DEU 2016 189170.0 CurrencyUSD 2 country/DEU 2017 308390.0 CurrencyUSD ``` Read `time_period` as text by setting `dtype`. If pandas reads it as an integer, `df["time_period"] == "2024"` returns no rows. ### Load it in R ```r library(arrow) library(dplyr) df <- read_parquet("nga-health/data/oecd-creditor-reporting-system.parquet") df |> filter(crsTypeOfFlow == "CRSTypeOfFlow10") |> group_by(time_period) |> summarise(value_musd = round(sum(value) / 1e6, 1)) |> tail(5) file.copy("nga-health/CITATION.cff", "nga-health-oda.CITATION.cff") ``` For the CSV, read `time_period` as text with `readr::read_csv("nga-health-csv/data/oecd-creditor-reporting-system.csv", col_types = readr::cols(time_period = "c"))`. ## Verify it worked The script creates `nga-health-oda.csv` and `nga-health-oda.CITATION.cff` in the folder where you run it. For the download named above, the 2024 row of `nga-health-oda.csv` reads `2024,843.9`. ## Troubleshooting - `ImportError: Unable to find a usable engine; tried using: 'pyarrow', 'fastparquet'.` pandas needs `pyarrow` to read Parquet. Install it, or run the script with `uv run --with pandas --with pyarrow python`. - `FileNotFoundError` on `nga-health/data/oecd-creditor-reporting-system.parquet`. Run the script from the folder containing `nga-health`, or use the `.csv` file if you downloaded the CSV. > [!NOTE] > Each source's licence is in `SOURCES.md` and on its page on [Sources](/data/sources). ## Next - [The data bundle](../data/bundle.md) lists every file in the zip and every field in `datapackage.json`, and explains how to get the same dataset again without a key. - The API quickstart fetches the same dataset directly, so you can skip the download.