Accuracy and bias in SparkToro’s data
Known biases in SparkToro’s dataset include:
- A bias to English-language profiles
- A bias to profiles located in the countries listed above
- A bias to people that use the web broadly, and, more specifically, the social networks listed above
- A bias to active social and web profiles (as we tend to look for those with real engagement and avoid bots, inactive, propaganda, and spam accounts)
Our data almost certainly includes other biases correlated with those above. For example, we likely over-index on wealthier households in English-language countries as these groups are more likely to have Internet access and regularly maintain/use social media accounts. We also inherently reflect the biases of societies and people, and do not attempt to alter our data or control for inherent bias. For example, SparkToro shows that the overwhelming majority of followed-accounts in fields like cosmetics, fashion, and style are run by women, while a similarly large majority of followed-accounts in fields like sports, computer science, and video gaming are run by men. These aren’t biases SparkToro’s founders support or want to re-inforce, but we also recognize that we cannot interfere in the data’s integrity without compromising its usefulness.
With regards to accuracy, SparkToro’s data tends to be very accurate when it is present. We do, however, occasionally encounter gaps in our ability to capture numbers, shares, content, or text from a social account or web profile. If you discover missing numbers/data in the tool, please email [email protected] and we’ll make sure to get it queued for re-crawling.