How to read a 0 to 100 reading

50 is an ordinary day

Every index on Boring News sits on the same scale from 0 to 100. The scale is not a percentage of anything and it is not a measure of danger. It answers one narrow question: compared with the recent past of the same measurement, where does today sit? A reading of 50 means today is in the middle of that range, an ordinary day. 0 is the calmest end and 100 the busiest.

Each index is built from several indicators, such as the number of strong earthquakes in the last seven days or how far a share price is below its recent high. Each indicator gets its own 0 to 100 score first. The index is then the weighted average of those scores, using the weights listed on How it works. So an index at 50 can mean every indicator is ordinary, or that some are busy while others are unusually quiet.

Percentile, not probability

The score is a percentile rank. In statistics, a percentile is a value with a given share of the data below it: about 70% of observations fall below the 70th percentile^[1]. Boring News works the other way round. It takes today's value and asks what share of earlier values were lower.

The rule, exactly as the site's code applies it: take the earlier observations of the same indicator, up to 365 of them. Count how many are lower than today's value. Add half of those that are exactly equal. Divide by the number of earlier observations and multiply by 100. Counting ties as half is a standard convention for percentile ranks^[2]. It stops a long run of identical values from all scoring 0 or all scoring 100. No score is given until an indicator has at least 20 earlier observations, so a new series cannot produce confident extremes from a few days of data.

A worked example. The React index counts earthquakes of magnitude 6 or stronger over the last seven days. Suppose this week had 5. Over the previous 365 days the seven-day total was lower than 5 on 300 days, exactly 5 on 30 days and higher on 35 days. The score is (300 + 15) ÷ 365 × 100, which is 86.

How long “the past year” really is

The site describes 50 as a normal day for the past year. That is exact for some indicators and approximate for others, because the comparison window is 365 observations rather than 365 calendar days, and some sources supply less history than that.

• Event counts (earthquakes, natural events, exploited software flaws, internet outages) have one value per calendar day, so 365 observations is one year.
• Exchange rates, oil prices and US Treasury yields are only published on working days, about 250 a year. Their 365 observations reach back roughly 17 months.
• The St. Louis Fed financial stress index is published weekly^[3]. The site fetches a little over two years of it, so each week is ranked against about the previous two years.
• The US share price series comes from a free data plan that returns about 100 trading days, so it compares against roughly the last four to five months.
• News-based indicators count headlines in GDELT’s published article lists. Their history builds forward from the lists the site has read, kept for up to a year, so for now they are ranked against fewer days than a full year.

For the news-based indicators, then, normal means normal for the last quarter. If a story has run for months, it becomes part of the baseline, and a news indicator can drift back towards 50 while coverage stays heavy.

Bands

To make the number quicker to scan, each index reading also gets a word.

• Quiet: below 35.
• Normal: from 35 up to 65.
• Elevated: from 65 up to 80.
• High: 80 and above.

The bands are fixed cut-offs chosen by Boring News, not statistical thresholds. A reading of 64 and one of 66 are almost the same, although one is called Normal and the other Elevated. Because an index averages several ranks, it rarely sits at the extremes. For an index to reach High, most of its heavily weighted indicators usually have to be unusual at the same time.

Change versus trend

Index pages show two kinds of movement, and they can disagree.

• Change is the reading today minus the reading 7, 30, 90 or 365 days earlier, in points. If there is no reading for that exact day, the nearest earlier one within three days is used.
• Trend compares today with 14 days ago. More than 5 points higher is rising, more than 5 points lower is falling, and anything in between is steady.

Suppose an index was at 62 two weeks ago, 48 a week ago and 60 today. The 7-day change is +12, while the trend is steady, because today is only 2 points below where it was 14 days ago. Both statements are correct; they compare with different days.

Changes are in points on the 0 to 100 scale, not percentages. A move from 40 to 60 is 20 points. It says the day moved from below the middle of its comparison period to above it. It does not say that anything grew by half.

Reliability and early readings

Each index has a reliability figure from 0 to 100%. It is the share of the index's weight that is backed by live data, with one adjustment: indicators built from news reports count at 60% of their weight, because a count of articles is not a measurement. Instruments and official statistics count in full.

Take the React index with all four indicators answering. Earthquakes (20% of the weight) and natural events (20%) count in full. Conflict reporting (35%) and sanctions and trade reporting (25%) count at 60%, adding 21 and 15. The total is 76%, which is labelled High. If the news source stops answering, reliability falls to 40%, which is labelled Low. The labels are High from 75%, Medium from 45% and Low below that.

An indicator counts as live only if its latest data point is no more than 10 days old. Older data is marked as stale and left out of reliability and drivers. When some indicators are missing, the index is the weighted average of those that remain, as long as they cover at least 40% of the total weight.

If even that much is missing, the site still shows a number rather than nothing: the average of whichever indicators did answer, marked as an early reading. Its history and change figures are worked out from those same indicators only, so they compare like with like. An early reading is a partial view and is best read as one.

What's moving it

Index pages list up to three drivers: the indicators pulling the reading furthest from 50. The pull of an indicator is its score minus 50, multiplied by its weight.

For example, in the Be Safe index, suppose hybrid incident reports score 90 and carry 30% of the weight. Their pull is (90 − 50) × 0.3 = +12. Exploited software flaws score 30 and carry 25%, a pull of (30 − 50) × 0.25 = −5. Hybrid incident reports are listed first, because they move the index more.

Drivers are the first place to look when a reading moves. A Be Safe reading pushed up by news reports means something different from one pushed up by measured internet outages. Measured, reported, counted once explains the difference.

Why a high score can be ordinary in a calm year

Because each score is a rank against the indicator's own recent history, it depends on what that history looked like. After a quiet year, a modest rise ranks high. After a busy year, the same value may rank in the middle.

Counts of rare events show this most clearly. Suppose the seven-day count of internet outages in one country was 0 on 330 of the last 365 days, 1 on 25 days and 2 on 10 days. A week with no outages scores 45. A week with one outage scores 94. A week with two scores 99. One event moves the indicator from the middle to near the top, because one event is unusual for that series, not because one outage is serious.

The same applies to financial stress. The St. Louis Fed builds its index so that zero represents normal conditions in financial markets^[3]. Boring News does not use that zero; it ranks the latest value against recent weeks. If stress has been below zero all year, a value that is still below zero, calmer than the Fed's own normal, can score above 50 here simply because it is higher than most recent weeks.

What the score cannot tell you

• How big something is. A rank removes the units. A score of 95 for earthquakes does not say whether that was 6 earthquakes in a week or 16. For scale, US Geological Survey figures for 2000 to 2021 average about 137 earthquakes of magnitude 6 to 6.9 a year, about 14 of magnitude 7 to 7.9 and about one of 8 or more^[4]. That is roughly three a week.
• Whether a rise in reporting reflects a rise in events. The news indicators count every matching English-language headline per day in GDELT's article lists, without dividing by the total amount of coverage that day. Syndicated copies, follow-ups and opinion pieces all count, and a busier news day in general can lift them. Grouping reports into single events happens only on the events list.
• What happens next. A percentile rank describes the recent past. It does not predict.
• Whether it matters to you. That depends on where you are and what you do. What to do when a reading is high covers this.
• Whether the inputs are the right ones. Combining several measures into one number is an established practice with its own literature: the OECD and the European Commission's Joint Research Centre publish a handbook of technical guidelines for building such composite indicators^[5]. Which measures go in, and with what weights, is a judgement. Boring News publishes both on How it works, and the weights in each index add up to 100%.

Read the score as a compact answer to one question: is this measurement busier or quieter than it has recently been? For anything more, follow the links on each index page to the underlying sources.