Democracy Vs Autocracy: Data Set & Method

This work involved significant data gathering & the drawing of boundaries in order to obtain useful insights. This post gives the details.

This project aims to determine if Liberal Democracy provides more stability than Autocracy based on historical data. Globally, we have seen significant growth in Liberal Democracy in the past two hundred years, growing from just a handful of regimes in 1800 toward around 50% (of varying quality) in the present day.

Therefore if we want to understand regime stability we need to compare like-for-like—as in comparing regime stability during a period where there are a decent number of democracies to compare against autocracies and how they interact with the global environment of this time.

Not just autocratic, but rainy too.

To do this we created a data set based on:

• All global regimes that existed from 1900 onward

If we are looking at regime stability, we need to look at survival rates. In 1900, there were a handful of regimes whose origins reached far back into history—and including them would distort the data if trying to understand modern-world trends. For example, The Kingdom of Siam, that started in 1238 and ended in 1932.

Likewise, there were countries ruled by old Empires; for example, the Ottoman or Russian empires, going back centuries and applying to multiple countries. Likewise, British and American Democracy (Bill of Rights 1689, Washington’s election 1789) are also outliers in this respect.

To reduce the distortion of these very old regimes, we used 1800 as a regime-start-date cut-off point. This helps us keep focused on regime performance in the modern period.

• Excluding regimes that started pre-1800 (Google Sheets)

The full data set included 736 global regimes that existed since 1900. Of these 46 started before 1800.

If we are looking at regime stability it becomes necessary to exclude Transitional Regimes that do actually transition because this is representative of stability, not regime failure. Clearly, some ‘transitional’ governments end up clinging to power and become just another autocratic regime. (Also excluded here; regimes that ended because the country was renamed or reformed)

• Excluding 28 Successful Transitional Regimes (Google Sheets)

After this, the total data set of regimes was: 662 (of 736 regimes gathered)

• Final Data Set (Google Sheets)

Data Gathering Technique

To gather data on every regime since 1900 was a fairly large task, especially when the context of the regime, its type, name, and what happened to it, was required. I used a mix of AI-supported data gathering and manual research to pull the data set together.

The process largely looked like this:

• Train a custom ChatGPT to execute a specific task set:

  • For a given state:

    • List all governments or regimes since 1900

    • Include:

      • Regime Name

      • Regime Start Date

      • Regime End Date

      • How it ended (e.g. military coup, revolution etc..) or other historical context

      • The regime that replaced it (used for continuity checking)

• Re-ask ChatGPT to check that no regimes were missed (this often brought up more data)

• Spot check Country Data

• Copy-paste to Excel & Repeat for each country

Note: due to the limitations of ChatGPT rate limits (or an appearance of AI laziness) I was only able to process one country at a time. Attempts to process more resulted in errors and omissions—quickly identifiable during spot checks.

Despite using AI tools to help the process, repeating the work for all countries took a considerable amount of time.

Assigning a regime as Autocratic or Democratic

Determining the nature of each regime was not completely straightforward either. Although many are fairly easy to determine as autocratic or democratic, others—for example, a presidential republic—may be ‘technically’ democratic but were, in reality, highly autocratic. An example might be modern-day Russia; technically a democracy but, actually, highly autocratic.

About 25 per cent of the data set needed manual research to determine the historically accepted nature of the regime.

Data Gathering Risks

The development of a custom Ai Mode, the double interrogation of ChatGPT, and the spot-checking of key data points reduce the chance of missing data and/or errors within the data.

That said, it is likely that there are some errors in the data set, and some eligible regimes are missing. Even accepting this, the data set is probably large enough for a small number of errors, or omissions, to have little impact on the overall results. This is especially the case when we are looking at developing big-picture views.

Future improvements

The result of this project has provided a firm justification for the value of democracy in terms of delivering political stability—and the benefits this brings. These are broad-brush findings that we consider would be highly indicative of any result that would otherwise be attained via more expensive/accurate data gathering or research methods.

If a higher degree of accuracy was required (for example, to reduce refutability) then the data-gathering process could be improved through the use of researcher time. If using an entirely manual research method, with skilled researchers, we would estimate such work to take around 500 to 600 hours—therefore coming at a cost (upward of) $25,000. If anyone wants to fund the production of a study that robust, get in touch.