How the pipeline works

1. S

   Setup

   Register each upstream data source — the named publisher, series id, vintage, and retrieval method — and prepare the working environment. Every table begins from a documented source of record (here, the U.S. Bureau of Economic Analysis Annual I-O Accounts).

2. L

   Loading

   Fetch and read the raw source data (BEA API pulls, archived spreadsheets) and check that the fetched units and dimensions match what the source promises before anything downstream runs.

3. P

   Processing

   Construction and transformation — and processing only. Assemble the Use and Supply tables, then derive the technical-coefficient matrix A, its square form, and the Leontief inverse L = (I − A)⁻¹, with a dimensional-analysis check whenever units differ.

4. V

   Validation

   Check each constructed matrix and series against the published BEA benchmarks — row/column totals, balance identities, unit and scale audits. Tables only pass when they reproduce the published values.

5. M

   Manual adjustment

   Apply and document any hand adjustment a source genuinely requires (for example a BEA redefinition or a one-off vintage fix). Each adjustment is recorded so the change is auditable, never silent.

6. A

   Analysis

   Compute the analytic quantities built on the matrices — output multipliers, backward/forward linkage indices, and the other I-O measures the studies and charts report.

7. O

   Output

   Write the publishable artifacts: the per-table CSV / Parquet files, the bulk archives, the catalog, the figures, and the documentation that ships with the data.

Reproduce a figure

You do not need the whole pipeline to check our numbers. An economy's output multiplier for a sector is simply the column sum of the Leontief inverse L = (I − A)⁻¹: the total output, across all sectors, required to deliver one unit of that sector's final demand. The transform below downloads the published L for 2024 (CSV), computes every sector's output multiplier, and ranks them — the same computation behind the Multipliers study. It is shown in both R and Python; switch with the toggle. Point L_url at any year, and change the column you sort on to explore.

# Output multipliers from the published Leontief inverse L = (I - A)^-1.
# The output multiplier of sector j is the column sum of L: total output
# (across all sectors) needed to deliver one unit of j's final demand.
#
# DATA:   point L_url at any year's Leontief inverse (the /api/table CSV).
# CHANGE: L_url (the year) and `n` (how many top sectors to print).

L_url <- "https://inputoutput.heterodata.org/api/table/2024/L?fmt=csv"
n      <- 10

# read the 71 x 71 inverse; first column is the sector index -> row names
L <- read.csv(L_url, check.names = FALSE, row.names = 1)
L <- as.matrix(L)

# output multiplier = column sum of L (one value per sector)
mult <- colSums(L)

out <- data.frame(
  sector     = names(mult),
  multiplier = as.numeric(mult),
  row.names  = NULL,
  stringsAsFactors = FALSE
)
out <- out[order(-out$multiplier), ]
out$rank <- seq_len(nrow(out))

cat(sprintf("economy mean output multiplier: %.3f\n", mean(out$multiplier)))
print(head(out[, c("rank", "sector", "multiplier")], n), row.names = FALSE)
# write.csv(out, "multipliers_2024.csv", row.names = FALSE)

# Output multipliers from the published Leontief inverse L = (I - A)^-1.
# The output multiplier of sector j is the column sum of L: total output
# (across all sectors) needed to deliver one unit of j's final demand.
#
# DATA:   point L_url at any year's Leontief inverse (the /api/table CSV).
# CHANGE: L_url (the year) and `n` (how many top sectors to print).
import pandas as pd

L_url = "https://inputoutput.heterodata.org/api/table/2024/L?fmt=csv"
n     = 10

# read the 71 x 71 inverse; first column is the sector index -> row index
L = pd.read_csv(L_url, index_col=0)

# output multiplier = column sum of L (one value per sector)
mult = L.sum(axis=0).rename("multiplier")
mult.index.name = "sector"

out = mult.reset_index().sort_values("multiplier", ascending=False).reset_index(drop=True)
out["rank"] = out.index + 1

print(f"economy mean output multiplier: {out['multiplier'].mean():.3f}")
print(out.loc[: n - 1, ["rank", "sector", "multiplier"]].to_string(index=False))
# out.to_csv("multipliers_2024.csv", index=False)

Both versions read the same published CSV and produce the same ranking — the studies ship full runnable bundles (R + Python + notebook + data) at /api/study/<slug>/bundle.zip.