Interpret your analysis, part 2: Advanced features

This article explores some more advanced features available to you as you interpret your event segmentation analyses. For a primer on the basics, review part one.

Rolling averages

Rolling averages display the unweighted mean, which smooths out a chart. This is useful if you have cyclical users—for example, people who use your product during the week, but not on weekends.

To apply a rolling average to your chart, click Advanced and select Rolling Average from the drop-down list.

This chart displays the daily event totals, without a rolling average. Notice the sharp peaks and valleys in the line.

But when you add a rolling average of seven days, those fluctuations disappear. That’s because each data point is now an average of the previous seven days’ worth of data.

Bear in mind that each day’s data is included in that day’s data point:

• When looking at the current day, Amplitude uses a dotted line to show data collection for today isn’t finished. Use an Offset in the date picker to exclude the current day from your analysis.
• Also, with a seven-day rolling average, the first six days of your selected time period fetch data from outside the selected time period. For example, in an analysis covering the month of February, the result for February 6th averages data over January 31st to February 6th.

Rolling windows

A rolling window is another method of smoothing out your data. It displays the aggregate last N days of information in a single data point. This is useful if you want to view aggregated metrics—such as your 7-day active user count—on a daily basis.

This differs from the rolling average, in that a rolling window doesn't average your data over the selected time frame. Instead, it sums the data.

To apply a rolling window to your chart, click Advanced and select Rolling Window from the drop-down list.

This chart displays event totals without a rolling window.

Below, the event totals display with a rolling window of seven days. Each day represents a sum of the previous seven days worth of events.

As with a rolling average, when using a seven-day rolling window, the first six days of your selected time frame fetch data from outside the selected time period. For example, in an analysis covering the month of February, the result for February 6th averages data over January 31st to February 6th.

Cohort filtering

When you select a rolling window, you can specify if aggregation happens before or after Amplitude filters results according to the cohort or cohorts you select.

• before cohort filter: Amplitude aggregates the event data, then applies the cohort and filters the results.
• after cohort filter: Amplitude applies the cohort and filters the results, then aggregrates the resulting event data.

Cumulative sum

Cumulative sum displays a running total of events in a single data point. For example, you might want to show a running total of revenue generated by purchase events. Cumulative sum helps you do that.

To apply a cumulative sum to your chart, click Advanced and select Cumulative from the drop-down list.

This chart shows a running total of purchases using the Complete Purchase event. The April 19th data point represents a sum of purchases on all the preceding days of the selected time frame. Here, that means April 5th to April 19th.

Using cumulative sum with uniques generate a count of unique users for each data point, with duplicates removed.

For example:

• On April 5th, User A triggered Complete Purchase.
• On April 11th, User A and User B triggered Complete Purchase.
• On April 19th, User C and User D triggered Complete Purchase.

On the data point for April 19th, Amplitude returns a total count of four because four unique users fired this event from April 5th to April 19th.

Real-time segmentation

View segmentation data in real time. However, there are some caveats:

• You can only segment one day's worth of data for real-time.
• Amplitude rounds event times down.
• Amplitude caches charts every five minutes for all users.

Period-over-period comparison

Use period-over-period comparison to compare the results of your current time range against previous time periods. The comparison control appears in the top-right of the chart, near the datepicker.

To start a comparison, click Previous Period vs. above the chart. This opens the Compare with panel.

Choose a comparison period

Select from the following preset comparison options:

Each preset displays the calculated date range so you can verify the comparison window.

Custom comparison periods

For more control, use the custom period options:

These options let you compare against specific dates, such as a product launch or campaign start.

Both custom options support a Rolling toggle. When you enable Rolling, the comparison period shifts forward automatically as time passes, keeping the same offset from the current date range. When Rolling is off, the comparison period stays fixed to the exact dates you chose, regardless of when you view the chart.

Compare multiple periods

You can compare up to two previous periods at the same time. Click Add Comparison in the Compare with panel to add a second comparison period. Each comparison period can use a different preset or custom date range.

To remove a comparison, click the X next to the period you want to remove.

View comparison results

Toggle between display modes in the Compare with panel:

• Absolute values — display the raw metric values for each period
• Percentage change — display the percentage difference between the current period and each comparison period

Period-over-period for custom formulas

Use period-over-period comparison with the custom formula metric. For example, compare your current rolling average with that of the previous month.