简体繁体 English

Google实验是否胜过？

[英]Google Experiments outperforming or not?

原文 2014-04-01 07:17:37 6 1 google-analytics/ ab-testing/ google-experiments

Today I met some strange thing in Google Analytics. 今天，我在Google Analytics（分析）中遇到了一些奇怪的事情。 One of my experiments ended with the probability of outperforming original of 94.7 % and outperforming original by 52 % and still with status "No winner found." 我的一项实验以94.7％的原始数据胜过原始结果的概率和52％的原始数据胜过原始结果的概率结束，但状态仍然为“未找到赢家”。

Has somebody met this situation? 有人遇到过这种情况吗？ Why is it so? 为什么会这样呢？

1 个解决方案

This means that while you do have versions that are better than your control, they aren't winning by enough to reach the 95% confidence interval . 这意味着，尽管您确实拥有比您的控件更好的版本，但它们的胜率不足以达到95％的置信区间。 In other words, you have a winner but it isn't statistically significant, or there's some chance that the outcome of this test is random. 换句话说，您有一个获胜者，但在统计上并不重要，或者该测试的结果很有可能是随机的。

Without seeing the numbers of impressions and conversions its difficult to be more specific. 如果没有看到展示次数和转化次数，就很难具体说明。 My guess is that either the conversion rate for this test is very low, the amount of improvement in the winning versions is very small, or both. 我的猜测是，此测试的转换率非常低，获胜版本的改进量很小，或两者兼而有之。

You have a couple of options for how to proceed. 您可以选择几种方法进行操作。

Repeat the test using only the control and the strongest variant. 仅使用对照和最强变体重复测试。 It's possible that by repeating the test with only two versions your test will reach significance. 通过仅使用两个版本重复测试，您的测试将很有意义。
Implement the strongest variant and find something else to test. 实施最强的变体并找到其他要测试的东西。 This is likely what I'd choose. 这可能是我会选择的。 Ideally you want to identify page elements that will have the greatest impact on conversion rates. 理想情况下，您要确定对转换率影响最大的页面元素。

Either way though, I would end the current test. 无论哪种方式，我都会结束当前的测试。 At this point there's nothing to be gained by letting it run longer. 在这一点上，让它运行更长的时间没有任何收获。