Skip to content

How to measure climate and gender finance with keyword markers

Tag a get_tossd frame with the twelve packaged keyword markers, then calculate disbursements per marker or combine multiple markers without double-counting activities tagged under several categories.

Steps

  1. Query data using the "analysis" column preset. Both extract_keywords and keyword_totals read keywords_raw, which is included in the "analysis" preset. The "minimal" preset omits it.

    import tossd_reader as tossd

df = tossd.get_tossd(years=2024, columns="analysis", units="usd_million")

  2. Extract marker columns for row-level analysis. The extract_keywords() method adds a boolean kw_<marker> column for each marker, including kw_gender, kw_adaptation, and kw_mitigation. Filter rows with a boolean mask to analyse matching records directly, such as breaking down marker disbursements by recipient.

    kw = df.tossd.extract_keywords()
gender = kw[kw.kw_gender]
gender.groupby("recipient_name", observed=True)["usd_disbursement"].sum().sort_values(
    ascending=False
).round(1).head(3)

    recipient_name
Developing countries, unspecified    9080.0
India                                7420.9
Indonesia                            3874.6
Name: usd_disbursement, dtype: float64

  3. Aggregate disbursements per marker with df.tossd.keyword_totals(markers=...). This method aggregates disbursements per marker without double-counting activities tagged for multiple categories. It recomputes marker masks from keywords_raw internally and appends a "Combined" row containing the union of all requested marker masks.

    df.tossd.keyword_totals(markers=["adaptation", "mitigation"])

           marker  usd_disbursement  n_rows
0  adaptation      38414.214244   41524
1  mitigation      61608.343767   34279
2    Combined      78643.461436   53822

    The individual marker rows sum to 100,022.6 million, exceeding "Combined"'s 78,643.5 million. An activity tagged for both adaptation and mitigation appears under each individual marker row, but only once in "Combined", providing an unduplicated total for climate finance.

    The n_rows column counts matching rows rather than distinct activities. Reporting providers may split a single activity across multiple rows with different keyword tags. Omit markers= to compute totals across all twelve packaged markers and the "Combined" summary.

Marker overlaps and vocabulary scope

The twelve markers operate as independent boolean flags without weights or partitioning, so marker totals overlap by design. A row reading False across all kw_ columns carries none of the packaged markers. In 2024, 10.9% of rows (51,436 of 474,026) contain keywords_raw tokens outside this twelve-marker vocabulary and remain False across every kw_ column.

Verify it worked

Sum the twelve marker counts and compare against the number of rows carrying at least one marker.

kw_cols = [c for c in kw.columns if c.startswith("kw_")]
print(kw[kw_cols].sum().sum())

258001

print(kw[kw_cols].any(axis=1).sum())

170206

The 258,001 marker instances span 170,206 rows. The difference represents rows carrying multiple markers, corresponding to the overlap accounted for by the "Combined" row in keyword_totals().

Troubleshooting

ValueError naming keywords_raw. Both extract_keywords and keyword_totals require keywords_raw. That column is included in the "analysis" and "all" presets. The "minimal" preset omits it.

tossd.get_tossd(years=2024, columns="minimal").tossd.keyword_totals()

ValueError: keyword_totals() needs column(s) keywords_raw, not present in df. Re-query with columns='analysis', or add keywords_raw to your columns= list.

ValueError naming a markers= value. keyword_totals rejects an unrecognised marker name and suggests the closest matches.

df.tossd.keyword_totals(markers="adaptaion")

ValueError: keyword_totals() marker 'adaptaion' not recognised; expected one of adaptation, biodiversity, covid_19, gender, idps_hostcommunities, mitigation, non_17_3_1, ppr_preparedness, ppr_response, refugees_hostcommunities, transnational_benefits_global, voluntaryrefugeereturn_reintegration. Closest matches: adaptation.

See also