# What was the Rogoff & Reinhart spreadsheet mistake?

In class our professor said that the Rogoff and Reinhart research on growth and debt was discredited due to spreadsheet mistake. What was this mistake? Also did they perform their research in excel instead of R, stata or some other advanced program?

• See here: peri.umass.edu/images/WP322.pdf Jul 24, 2020 at 9:39
• @MichaelGreinecker you should expand it into an answer
– 1muflon1
Jul 24, 2020 at 9:49
• @1muflon1 I leave that to those with more motivation. Jul 24, 2020 at 9:50

Reinhart and Rogoff (2010, RR) claimed that GDP growth is much lower when debt/GDP > 90%.

Herndon, Ash, and Pollin (2014, PDF, HAP) placed their errors into four categories:

1. Spreadsheet error: RR omitted Australia, Austria, Belgium, Canada and Denmark from the analysis (apparently a coding error somehow dropped the first five countries when sorted alphabetically).
2. One small transcription error: RR incorrectly transcribed a figure of "-7.6%" in one spreadsheet to "-7.9%" to another.
3. Inappropriate weighting: Example: New Zealand had only one year (1951) with debt/GDP > 90% and in 1951, its GDP growth was -7.6%. The UK had 19 years with debt/GDP > 90% and in those 19 years, its average GDP growth was 2.4%. The usual practice would be to count New Zealand 1951 as one country-year observation and those 19 UK years as 19 country-year observations. But instead, RR averaged the 19 UK years as one single observation that carried the same weight as that single New Zealand 1951 observation.
4. Selective exclusion of available data: Example. RR exclude Australia (1946–50), New Zealand (1946–49), and Canada (1946–50) without giving any explicit reason. But these country-year observations are precisely instances of high GDP growth and debt/GDP ratios.

Errors #1 and #2 are clear cut (and may be fairly characterized as "Excel spreadsheet errors").

Errors #3 and #4 are a little more subjective.

(Neither RR nor HAP state the program used. But as reported in e.g. NYMag, following repeated requests from Herndon, Reinhart emailed to Herndon an Excel file and it was in this file that Herndon discovered errors.)

The paper by Herndon et al. shows that there were "coding errors, selective exclusion of available data, and unconventional weighting of summary statistics". So it's a little deeper than a mere "spreadsheet error" and isn't software-specific, but the term makes good press. Herndon was interviewed on the Colbert Report and I believe that may be where the paper got popularly associated with that term.