Website traffic data typically looks like this:
Ups and downs, peaks and valleys.
If we’re doing our job well, we typically expect traffic to trend upward over time, but in any given month, it’s hard to tell whether peaks or valleys are worth paying attention to.
Did we do something great and trigger a new phase of growth? Do we benefit from the new Google update? Or is this just normal variation, part of the natural ebb and flow of people discovering our site?
Or suppose you make changes to your content flow – you Trim and redirect A bunch of old content, and then the traffic drops off the next month. Is that dripping? Caused Is it due to change, or just coincidence?
I’ve been trying out a simple statistical tool designed to help answer the following questions: XMR Chartalso called a process control chart.
Here is the XMR chart:
XMR charts are designed to tell you whether any single data point in a time series is likely to be caused by normal fluctuations (“regular changes”), or is a sign that something is happening and needs to be investigated (“unusual changes”).
The XmR chart consists of the x valuethe “things” we care about—such as widgets produced or sales closed)…
…as well as MR diagrams (with Moving rangebasically the “gap” between each data point):
In its simplest use, if you plot your data on a chart and it swings up and down around the center line without crossing the upper and lower limits – no problem! These ups and downs may represent normal changes.
But any point that occurs outside the upper or lower bounds (as shown in the picture) Red) should be considered an anomaly requiring investigation.
In the X chart above, the time series appears to show regular changes until the first red out-of-bounds point on January 16th.
XMR chart shows something What happened on the 16th disrupted our production flow (for better or worse). Our job is to investigate the cause.
Side note.
The middle line is the mean of the data set; the upper and lower bounds represent 3 standard deviations from the mean (called three sigma). Any points outside these upper and lower bounds are likely to be anomalies and not part of the original probability distribution.
The XmR chart can also show you other “signs” (e.g. eight consecutive points on one side of the average line representing another abnormal change), but I’ll leave it to you to investigate these signals.
When I started reading XMR charts, I thought of an obvious use: identifying the impact of Google’s algorithm updates.
If a site’s traffic drops to zero, it’s easy to say “we got a manual penalty.” But for smaller changes, such as a drop in traffic for several consecutive months, it can be difficult to pinpoint the cause. Have we been caught by a Google update? Is it seasonal? Or is this just a coincidence and traffic may return to normal in the future?
The following is the monthly organic traffic data of Ahrefs blog for two years, taken from website browser And plotted on the XMR chart:
Now…this isn’t particularly useful.
There are a large number of data points outside the expected range (red) and very few data points located closer to the center line than the quartile limit (orange).
XMR plots should show abnormal changes in a consistent process, but in this image, almost all data points indicate abnormal changes. What gives?
Flowcharts are designed around simple manufacturing processes and they are very effective when the expected output of the process is constant.
If your goal is to produce 10,000 widgets per week, the XmR chart will help you determine whether producing 5,600 widgets per month is a normal “phenomenon” in daily operations, or whether it is caused by a real problem that needs to be investigated.
Website traffic is more complicated. There are many variables that affect traffic:
- fluctuating search volume per topic,
- personal ranking position,
- New competition articles,
- Search function,
- Seasonal,
- Release frequency,
- Google algorithm update…
This means that performing XMR analysis on a long stream of data may not be very helpful. Your “blog flow” is unlikely to remain stable for long.
As far as I’m concerned, this particular two-year snapshot of data may not come from a single stable process, and there may be multiple probability distributions underlying it.
But we can make analytics more useful.
Best practice with XmR charts is to limit analysis to a period of time when you know the process is relatively static, and recalculate when you suspect something has changed.
See the moving range chart for this data below, with large traffic differences occurring in November and December. We should investigate possible causes.
I know our publishing frequency is fairly static (we’re definitely not doubling our content output). Seasonality can cause traffic to drop, not surge (we’re writing about SEO, not holiday gift guides).
But Google made a major update in early December:
If we assume something happened For our blog flow at this time (perhaps due to traffic changes caused by Google updates), we can add a new separator line to the XmR chart.
Instead of trying to analyze the traffic as a single process, we can treat it as two processes and calculate the XmR chart separately:
Now the first process looks stable (full of black spots). The second process also shows less extreme changes (red), but still too much moderate change (orange) to appear stable. There may be another process lurking in there.
and every Rule of thumb For analyzing XMR charts: “When ‘long-term’ data remains above or below the average, the XMR chart needs to be revisited for duration.” This trend started in late summer (which is also Around the time Google announced another core update):
We can add another delimiter at the beginning of this “long-term” data to create three separate XmR analyses:
Doing so, all three analyzes appear to be stable, with no points of extreme variability. In other words, we seem to be doing a pretty good job of capturing three different processes occurs in our traffic data.
From this analysis, it is likely that our traffic was affected by external factors during the two major Google updates.
Now…this is basically an exercise in data torture after the fact. We cannot infer any causal relationship from this analysis, and it is entirely possible that other arbitrary divisions could produce similar results.
But it doesn’t matter. These charts can’t give you a clear, concrete reason why Why Your traffic changes, but they can tell you where to seeand help you determine whether addressing traffic dips or spikes is the best use of your time.
The ultimate measure of a model’s usefulness is its ability to help you predict thing. Will XmR charts help me run my Ahrefs blog better in the future?
I think so.
Assuming that my “blogging process” remains relatively stable – I post at the same frequency, target the same topics, and compete with the same competitors – I now have a “stable” set of data that I can use to inform future plans. Traffic data provides additional context:
Over the next few months, I can figure out whether drops or spikes in traffic might be the result of normal variability, or if there’s a change that needs my attention (like a Google update).
For example, if this happens to my traffic next month…
…I know that given this distribution, the drop in traffic is probably a normal and exciting change.
But if you do…
… There may be something else at play.
For extreme traffic changes, you can usually “look at” the traffic graph and guess what’s going on. But XmR charts are useful for more subtle changes, and I’ve had the opportunity to be able to identify and act on a month’s worth of data. so cool.
final thoughts
Tackling traffic variability is a huge challenge for SEOs and content marketers (and we’re working on several ways to help you identify the signal among the noise of traffic data).
In the meantime, I’ve found the XmR chart to be an interesting tool to have in my toolkit, helping to put my monthly reporting numbers into context and demonstrate when I should (or shouldn’t) be expending effort troubleshooting Month question.
(At least, when your VP sends you a rude email at 3 a.m. complaining that traffic dropped 8% last month, the XmR chart might give you enough confidence to say “stop bothering me.” )
Side note.
Thanks Benjamin EliasThe VP of Marketing at Podia introduced me to the XmR chart.
