IP90 and the Datasaurus: The dangers of summary statistics

June 13, 2017 by

I recently read the article Same Stats, Different Graphs which spoke about Anscome’s Quartet, four highly varied datasets that are identical when examined using summary statistics, then took it to the next level with the Datasaurus Dozen by demonstrating how highly varied datasets could produce the same summary statistics.

Datasaurus Datasource: Alberta Cairo 

The article states “It can be difficult to demonstrate the importance of data visualization. Some people are of the impression that charts are simply “pretty pictures”, while all of the important information can be divined through statistical analysis.” It also references Alberto Cairo who created the Datasaurus dataset to urge people to “never trust summary statistics alone; always visualize your data” because visualization can reveal valuable insights that could be otherwise missed.

This inspired me to revisit my 2015 blog How useful are IP30, IP60, IP90 … initial production measures where I illustrated how two wells with identical IP90 production performance measures had very different production profiles. 

To further illustrate the dangers of using near-term production performance measures like IP90, without visualizing the production, I’ve expanded the dataset from my previous blog to show just how different the production profiles are of 33 Montney Regional Heritage wells that have the same IP90 values (i.e. IP90 values between 4400 and 4500 mcf/day). The chart on the left shows the rate vs. time profiles of each well, while the chart on the right shows the cumulative production vs. time.

IP90 is the average daily production rate after 2160 hours of production. In our sample set of 33 wells it takes from 3 to 22 months to produce for 2160 hours, and the cumulative production after 24 months ranges more than 5 fold, from 516 mmcf to 2,693 mmcf. This further demonstrates that IP90 is not a reliable indicator of longer term production. Also, the need to visualize the production profile is becoming increasingly important as rate restriction becomes a more common operational practice (look for wells without an obvious decline profile in the chart on the left).

IP90 is sometimes used as a proxy for the long term production potential of a well (i.e. Estimated Ultimate Recoverable or EUR). It’s important to note that the uncertainty of any predictive measure decreases as you include more production history. The challenge is often to find the balance of including enough history to have reasonable certainty, but not requiring too much history so more recent wells are excluded from you analogue selection. The table below is useful in deciding how much history you should use to reduce the uncertainty of you production performance measure (full details of this correlation analysis can be found in my previous blog).

Conclusions:

  1. More data is better
  2. Don’t rely on summary statistics alone
  3. Visualize your data, from multiple perspectives, to elevate your understanding
Next Post
  • Marcel Preteau says:

    Hey Bertrand,
    You provide a very useful caution and “sound bite”. Given the pace of technological change, no prudent (and active) operator can just wait on their completion design decisions until they have achieved 100% certainty. But how much time is enough? We have proven to ourselves that IP90 is a weak indicator in assessing the likely ultimate performance of a well. IP180 is much stronger…but is it enough to shape changes to very significant completion designs? I suppose that time will tell.

    • Jeanne Phené says:

      Great article, I agree with all comments above. However, executives, operators and folks optimizing completions are desperate to have a pragmatic early indicator of success and that has been IP90, it is in our DNA. Any data set variability is commonly dismissed as ‘rocks’. The math to incorporate variability we see in the graph on the left above into our forecasts and plans is a bit beyond the intro to stats we ALL took. Unfortunately IP90 is often extended to EUR ranking often with NO attempt at bringing decline forecasting into the ongoing decision process, we’ve raised a ton of capital with such stuff! This results in a whole lot of uneconomic production coming in to market [I can almost get any IP90 I want with a big enough frac and I will when I am desperate to meet a boed commitment] and then what happens…a bit of loose money all round and a piddly 5 mmbld floods the market with our product and price fails. Not news but hard to get through to the poor folks up to their necks in alligators.

Leave a Reply

Your email address will not be published. Required fields are marked *

Learn how VERDAZO can enable your business

Contact us
Product Industries Services Clients Learn Contact
Blog News About Careers

Searching VERDAZO