Don't let anything slow you down
All of our phone and SIM Only plans come with blazingly fast 4G speeds, so you can browse, chat, stream, game and stay connected on the go.
All on a network that’s faster on average than O2, Vodafone, Three, and Sky!
About Speedtest Intelligence® Data Methodology
With millions of tests taken each day, Speedtest has the most comprehensive view of worldwide internet performance. We have worked diligently to devise the most accurate method for determining who are the fastest ISPs and mobile networks. We believe we have constructed an unbiased and equitable approach that ensures stable results, controls for extraneous variables, and ensures every internet user gets a fair number of “votes”.
Methodology Overview Methodology Overview
Our methodology across platforms (web, mobile, etc.) begins with a raw data test and, after a series of filters, we aggregate the data to provide a final set of results. During the filtering process, we believe we have constructed a fair and balanced approach for producing network metrics that ensures stable results, controls for extraneous variables, and ensures every internet user gets a fair number of “votes”.
The raw test data are the raw results taken by Speedtest users from each individual platform.
Mobile results include Android and iOS results taken using their respective Speedtest application. Each mobile result comes parceled with data covering several categories of interest for constructing samples for the Awards rankings and Intelligence metrics, such as MNC/MCC codes, GPS location, connection types, download speed, upload speed, and latency, among others. We rely mostly on GPS coordinates to get a latitude and longitude coordinate for a device at the time a test was taken. However, in some cases we might have to fall back on GeoIP systems to determine location.
Sample Construction Sample Construction
“Sample Construction” enables us to create standardized data points that can be used for further statistical analysis. In general, the goal is to reduce the effect of extraneous variables, and ensure every user gets fair representation in the final analysis, as well as to prevent gaming of our ranking systems. In order to control for variables, as well as ensure each user gets fair representation in the data, we must construct samples. Otherwise, for example, a power user that takes 100 times as many tests as a casual user would have their correlated speeds weighted higher in many of our final aggregates.
A single “sample” is constructed on our end straight from a database table containing raw test results by using a query language, and each sample is considered an aggregate. That means we group our data by several key identifiers and produce a few fields that are aggregate values, such as the mean download speed, or mean upload speed for those key identifiers. In practice we are producing user-samples defined as “mean speed for user/device X per time period, in location Y, under provider Z”.
During the aggregation step, we use the user-samples constructed to create data subsets and compute various statistics on these subsets that can be used for tracking progress or comparing providers and/or locations to one another. The thresholding and statistics used are designed to capture some attribute or information that can be used for comparing these data subsets to one another.
We aggregate the samples by time frame, provider and location using statistics that best represent the attributes we are trying to measure.