Insights Stream estimation (Legacy)
This article contains information about Insights Stream estimation.
Why are my metrics being projected/estimated?
Whenever there is a need to estimate or project at least one of the metrics for a search stream, we will display the reason at the top of the page in a message like the one below.
When do metrics need to be estimated?
Metrics can be estimated for search streams when one or both of the following occur:
- When the volume of mentions for the stream is very high. (Projection)
- When complex filter combinations are applied to the stream. (Estimation)
For Insights search streams that gather a very high volume of mentions (tens of thousands or more per day), it is not possible to retrieve every single mention. We retrieve as many mentions as possible, and then based on the volume and the rate at which the mentions come in, the actual number of mentions is projected.
Projected metrics are statistically accurate, but may not exactly match the actual number of mentions.
Please note that such high-volume streams are few. Making search expressions as accurate as possible will in most cases generate streams well below the threshold for projection.
Metrics in Hootsuite Insights are computed by counting each property of a mention individually, and one count is maintained for each filter that we support. For example, as a tweet’s language, sentiment and location are determined, each of these properties is added to the count for that respective filter (e.g. sentiment positive, sentiment negative or sentiment neutral).
You can then filter for all mentions with positive sentiment, or all mentions in English, and those counts will be exact and accurate. However, if you filter for all mentions with a positive sentiment whose text is in English, we couldn’t provide this number exactly, unless we maintained a third count specifically for mentions with a positive sentiment that are in English.
As it is not possible to maintain a count for every possible combination of the filters we support, we need to estimate some of those combinations. Even though we might not have a specific count for all complex filter combinations, we can determine how many mentions that filter combination has by doing a few computations. Depending on the exact filters, the estimation can have a high degree of certainty.
Do more filters mean less accurate results?
No. Narrowing search results with more filters doesn’t mean the accuracy of the metrics will decrease. In some cases the opposite happens. If the applied filters return a small number of mentions, in many cases we can go through every single mention and build the complex count combination on the fly, giving 100% accurate results.
What are the consequences of metrics being estimated?
The main consequence of estimation is that the numbers can have a degree of uncertainty. Even though the metrics are statistically correct, they may not exactly match the actual number of mentions. For example, the metrics could show that there are 10 mentions in English with a positive sentiment published from New York, but the mentions section may contain 11.