Comparing Internet Access Speed on Between Ebay and Amazon on Black FridayEssay Preview: Comparing Internet Access Speed on Between Ebay and Amazon on Black FridayReport this essayStatistics classComparing internet access speed on between ebay and amazon on black fridayIntroductionWhy is it relevant?In a world where time is money, that is especially true in the internet retail business, milliseconds can cost companies millions of dollars. A report from Amazon.com states that for every 100 milliseconds that it takes for the website to load, it will cost the company 1% of revenue (Kohavi & Longbotham, 2007). That is a large chunk of money for high volume sites internal retailers to miss out on.
In our research we focused on one of the biggest online retail day of the year, Cyber Monday. Its the Monday after Thanksgiving when Americans are shopping for the holidays so there are high number of users. We focused on eBay and Amazon, which are two of the leading online retail sites in the U.S., and their associated website load time. In the following pages, we will take you through the journey on what we analyzed and how we came to our conclusions.
Internet Performance – Game 7 of the World SeriesOur original idea was to take the load times of social website and see if cultural events could slow down the performance. We took the information from Twitter and Facebook over the same period of time as Game 7 of the World Series between the St. Louis Cardinals and the Texas Rangers. This game had the highest TV rating for any baseball game in seven years, with an estimated 25.4 million people across the U.S. tuning in (Singer, 2011). We got great data for the event but we were unable to come up with the play-by-play in real time to compare when significant events in the game occurred. As a result, we re-focused our attention to the Amazon and eBay load time comparison.
Data AnalysisInternet PerformanceWe looked for indicators of the online shopping experience during a high-traffic event, namely Cyber Monday. Specifically we looked at indications of slow site response time experienced by users engaged in online shopping. We compared data collected from 11/21/11 (the Monday prior to Cyber Monday), and 11/28/11, Cyber Monday. Our ultimate goal is to provide information to retailers so that technology decisions can be made to provide the best online shopping experience to their customers.
Amazon vs. eBayOur sample size was over 50,000 data points which provided the information regarding internet connection data for both November 21, 2011 and November 28, 2011. The 21st is the Monday before Black Friday and the 28th is Cyber Monday. The data was evaluated and analyzed from both days and complied the connection rate and connection speed for each website.
Both days were consistent with each other in our findings, with Amazon.com having a higher connection rate than eBay. On 11/21, Amazon outperformed eBay by 1.8% and on 11/28 they outperformed by 2.06%. The connection rate was almost identical from day to day, Amazon was in the high 96% and eBay was in the high 94%.
November 21, 2011 (12:00am – 2:58pm PDT)Connection RateSuccessfulTotalPercentageAmazon.com313813245896.68%eBay.com201022118694.88%Connection SpeedNumberSpeedAverageAmazon.com3138120600986656.48eBay.com2010214883936740.42November 28, 2011 (12:00am – 2:58pm PDT)Connection RateSuccessfulTotalPercentageAmazon.com326513371496.85%eBay.com210542221194.79%Connection SpeedNumberSpeedAverageAmazon.com3265121814207668.10eBay.com2105415849100752.78The most revealing information discovered, was the connection speed and the relationship to the connection rate. When I compiled the connection speed of each site, eBay took 84 milliseconds longer to load than Amazon. What we interpreted is that the website that took longer to load (eBay) had a lower connection rate.
Sample Mean ComparisonIn order to confirm that there is significant evidence to say that the means of Amazon and eBay are different, we ran a t-test with the data from November 21st. We used a level of significance of 95% and used a two-tailed test. The sample size was so large that it came close to representing the population. As expected the t-test statistic came out with a value of -49.117 and the lower critical value is -1.96 so we can confidently reject the null hypothesis.
Pooled Variance t-TestHypothesized DifferenceLevel of SignificanceAmazon.com SampleSample Size31381Sample Mean656.480Sample Standard Deviation199.460eBay.com SampleSample Size20102Sample Mean740.421Sample Standard Deviation171.888Intermediate CalculationsAmazon.com Degrees of Freedom31380eBay.com Degrees of Freedom20101Total Degrees of Freedom51481Pooled Variance35786.474Standard Error1.709Difference in Sample Means-83.941t Test Statistic-49.117Two-Tail TestLower Critical Value-1.960Upper Critical Value1.960p-value0.000Reject the null hypothesisDate Mean ComparisonThe next hypothesis we wanted to test was if the difference in connection speed from November 21st and November 28th was significant. We combined both sites data and
s-test data. The tests were conducted with eBay, the only other large online e-Bay site (I’ve never done tests for eBay), and found the correlation to the difference in connection speed test was 2.8. We chose the test because it had a more robust and reproducible design with respect to comparing over 100 sites. We also decided to test it with other tests and compared the correlation coefficients to obtain better results. We used this method to perform the above three tests. The test showed no significant p-values but the correlation with the 2.8 difference was nearly twice as high: <10e-eBay% of difference. And the other two test were highly significant: <5% of difference. The following table shows the correlation between the tests and the data. We first use an i-test for any set of tests, then use the test's logistic parameter, and finally for any set of 3 or more tests, and we then use the log and n test (the last 2 tests with at least a single logistic parameter; which are the tests used. We also include our test's standard deviation into the final set of data. By using this method, the following table shows and compares the 3 different test effects for comparing a 2,000 base sample size and the results for two different tests. It is worth noting that the 3 tests that produce the strongest and best correlation seem to be those tests that use logistic regression as the standard parameter, whereas the other tests in the table are those tests based on standard errors in two different sets of tests and hence which are used within the logist-query test. Because there are so many different logistic regression methods as well as the fact that there are few logistic regression techniques, this is also a good point to try using. The data is available through https://www.i-data.net/test-data-logistical-relationses/ and the results are available from my blog http://blog.i-data.net/ and here:http://blog.i-data.net/post/271735.html (If you think you know the data from my last blog you should check that it is up-to date for the second blog post, the first is http://blog.i-data.net/blogs/2011/08/15/the-data-is-up-to-date-for-the-second-blog-post.html ). Some of the graphs used in this document are drawn randomly (using the standard error statistic as the baseline, to see if they could be included in the same regression model) and it is important to observe that only a small part of all the