Google Analytics Graph Scaling & Time Series Analysis using Tableau

An analytics tool should present data in a fairly unbiased way. The subjectivity comes into play once a human begins interpreting the data. Beware when using the new Google Analytics. Their scaling algorithm is not quite ‘stable’, case in point:

The above data is exactly the same, the date granularity chosen gives rise to differing trends:

1st Chart: Calendar Q1, daily visit data.

2nd Chart: Calendar Q1, weekly visit data.

Without the benefit of viewing the scaling on the graphs you’d think traffic grew tremendously over the period in the weekly case, but it was basically flat! Well, it’s trending up ever so slightly…don’t make me generate a random series in Excel:

(below) – that’s definitely going up, and it’s definitely random! I bet you could convince quite a few people there’s an uptrend at work here…it’s about to hit a new high, after all. It may be true, but without the benefit of observing prior random patterns (not shown) how could you know? Given that it is random, it’s due to decline soon, but given the pattern of a high followed by yet another high…the optimist might sense a breakout coming.

The issues only grow when analyzing dual axis charts: you have to consider whether the axis are synchronized. This really means if you move up half the left axis, the % change in the metric is the same as the % change if you moved the same distance on the right axis.

Time Series Analysis Using Tableau – Axis Synchronization

Above – I’ve generated 2 random series with daily data for 1 year. The 2nd series is simply a constant factor of the first (eg Xc or something multiplied by .75, always .75 each time). The first chart has NON synchronized axis – you can see that values ascend more in a non-uniform way. However, if you ask the analytics tool to synchronize the axis, the 25% discount the blue series has applied to it becomes obvious. They are perfectly correlated, so much so that without adjusting the axis they’d be right on top of one another.

When you’re dealing with metrics that are on different orders of magnitude, say: revenue and conversion; this isn’t feasible because conversion would be well under 1.0 while revenue would be in the 1000′s or tens of thousands or millions, etc…

Bottom line – while it’s nice to visually interpret data, be very aware of not only the direction of the change, but the magnitude as well.

Hope this post was off the charts,

Jeff