These kinds of quirks in retention figures are typically down to different cohorts being different sizes and showing different behaviour. You can get more details on your data by switching from Retention view to Change over time, using the dropdown options below Retention View / Usage Interval View switch.

For more details and some examples, check this recent thread:

To add to MikkoKarvonen’s point, it is important to recognize how your chart’s interval affects the percentages displayed. The reason why the percentages for the ‘Last 8 weeks’ chart are higher is due to the fact that more time has elapsed which gives more users a chance to be week 1 retained. 

Let’s look at an example from our demo to illustrate this point by examining the breakdown table: Example

If you expand the breakdown table below your chart, you will see that there are ​several weeks worth of data. The figures in the top row represent the deduplicated weighted averages for weekly retention which is reflected on the chart, while the numbers in the columns below it were used to calculate that average. Essentially, each data point for week N will be the weighted average of the percentage of users retained that week. 

In this chart, week 0 represents the week of June 21st, June 28th, July 5th, July 12th, and July 19th. Consider the cohort of 62,786 users who entered your retention analysis in the week of June 28th. Their week 0 occurred sometime in the week of June 28th, their week 1 occurred sometime in the month of July 5th, and so on. 

When you change your chart’s interval from 4 weeks -> 8 weeks, this allows for more week 1 data to be collected and calculated into the weighted average for week 1 retention.