標記偏見

“Beware of the HiPPO in the room” — The risks and dangers of top-down, intuition-based decision making are well known in the business world. Experimentation and data-based decision making become widely acknowledged as the right way to steer a business.

“當心機房中的HiPPO” —自上而下，基于直覺的決策制定的風險和危險在商業(yè)界眾所周知。 實驗和基于數(shù)據(jù)的決策被公認為是指導業(yè)務的正確方法。

For a good reason: Leading experimenters such as Netflix, Google and Booking show that making decisions based on facts and evidence rather than intuition can lead to exceptional business success.

有一個很好的理由：Netflix，Google和Booking等領先的實驗者表明，根據(jù)事實和證據(jù)而不是憑直覺做出決策可以帶來非凡的業(yè)務成功。

But what if in the course of this development the HiPPO (Highest Paid Person’s Opinion) is not the one to be afraid of anymore? What if the person that should help to fight top-down decision making took his place?

但是，如果在這種發(fā)展過程中，HiPPO(最高付費人士的意見)不再是一個令人恐懼的東西呢？ 如果應該幫助反對自上而下的決策的人接任該怎么辦？

What if the analyst is the new biasing factor in decision making?

如果分析師是決策中的新偏見因素怎么辦？

Let me be clear. Having personal opinions about new ideas, suggestions and approaches couldn’t be more natural. We all have our cognitive biases. But we can not ignore this fact just because of the allegedly safe framework of evidence-based decision making.

讓我清楚一點。 對新想法，建議和方法有個人見解不會再自然。 我們都有我們的認知偏見。 但是，我們不能僅僅因為所謂的基于證據(jù)的決策安全框架而忽視這一事實。

UnsplashUnsplash

偏差分析結果 (Biasing Analytics Results)

The analyst’s responsibility is to paint a clear picture of the business’ situation and inform decision-making. And there are plenty of ways how an analyst can, willingly or entirely unaware, bias the final decision that is made. I like to split those into conscious and unconscious biases.

分析師的責任是清楚地描述業(yè)務情況并為決策提供依據(jù)。 分析師可以通過多種方式自愿或完全不了解最終決策。 我喜歡將這些分為有意識的和無意識的偏見。

自覺偏見 (Conscious Biases)

Let’s have a look at the first kind, where analysts make deliberate decisions that will impact their results. Conscious biases are closely connected to the analyst’s personal opinion. This can be about a new marketing campaign or a new product feature. Whether the analyst believes in a specific idea can significantly impact how the following research is being conducted. And if it is conducted at all. Let’s have a look at a few potential sources for conscious biases:

讓我們看一下第一種類型，在這種類型中，分析師做出會影響其結果的深思熟慮的決策。 有意識的偏見與分析師的個人看法緊密相關。 這可能與新的營銷活動或新的產(chǎn)品功能有關。 分析人員是否相信特定想法會嚴重影響以下研究的進行方式。 如果是進行的話。 讓我們看一下一些有意識的偏見的潛在來源：

“Can you just give us a rough estimate for this particular metric?”

“您能給我們這個特定指標的大概估算嗎？”

a) Making Guesses. Questions for estimates and opinions are more or less an invitation for introducing biases. Obviously you can’t know all the numbers by heart and the best thing to do would be to go back to your desk, check the metrics and report them back. But checking every single metric costs too much time. Often enough, we simply trust our intuition, which strongly correlates with our personal opinion about a specific idea. So you make an educated guess about what the number might be. At this point, analysts can already substantially impact whether an initiative is pursued and what everybody’s expectations are. The first number one comes up with serves as an anchor figure and sets expectations stakeholders might reference in the future to assess an idea’s potential.

a)猜測。 有關估計和意見的問題或多或少地引起了人們引入偏見。 顯然，您不能一味地知道所有數(shù)字，而最好的辦法是回到辦公桌前，查看指標??并將其報告回來。 但是檢查每個指標會花費太多時間。 通常，我們只是相信我們的直覺，這與我們對特定想法的個人看法緊密相關。 因此，您可以對數(shù)字可能進行合理的猜測。 在這一點上，分析人員已經(jīng)可以對是否采取主動行動以及每個人的期望產(chǎn)生實質性影響。 第一個數(shù)字一個想出了作為錨人物和套的預期利益相關者在未來可能會引用到評估一個想法的潛力。

“Traffic is so low on this page, it’s not worth looking further into this.”

“此頁面上的流量如此之低，因此不值得進一步研究。”

b) Giving Personal Opinions. Sometimes we might be tempted to provide no number at all and instead give a personal opinion. While this opinion is (hopefully) based on facts and the analyst’s experience, it can still strongly correlate with one’s subjective opinion about the idea discussed.

b)發(fā)表個人意見。 有時我們可能會不愿提供任何電話號碼，而是發(fā)表個人意見。 盡管這種觀點(希望)基于事實和分析師的經(jīng)驗，但仍可以與人們對所討論想法的主觀觀點密切相關。

c) Depth of Research. After kicking off the research, the question is:

c)研究深度。 在開始研究之后，問題是：

When does an analyst have enough information to give a good recommendation or overview for a particular problem?

分析師何時有足夠的信息為特定問題給出好的建議或概述？

Of course, you can always drill deeper into a specific topic to get more evidence to support a decision-making process. Analysts might be inclined to dig deeper into an area to prove or disprove a particular idea they have a strong opinion about. Simultaneously, we might put less effort into a research question where the outcome is expected to be less exciting or the decision that has to be made seems to be pretty trivial anyway.

當然，您總是可以更深入地研究特定主題，以獲取更多證據(jù)來支持決策過程。 分析師可能傾向于更深入地研究某個領域，以證明或反對他們有強烈看法的特定想法。 同時，我們可能會在研究問題上投入更少的精力，因為預期結果不會那么令人興奮，或者必須做出的決定似乎微不足道。

d) Setting targets. The analytics and experimentation landscape itself invites analysts and anybody who operates in it to introduce biases at some points. Be it setting the right significance or power level for an AB-test or selecting an appropriate metric to measure a new feature or a campaign’s success? Those are, to a certain degree, subjective decisions the analyst has to make to produce any results. But at the same time, those can have a significant effect on the actual outcome of the research.

d)設定目標。 分析和實驗環(huán)境本身會邀請分析師和其中的任何人在某些時候引入偏見。 是為AB測試設置正確的重要性或功率級別，還是選擇適當?shù)亩攘繕藴蕘砗饬啃鹿δ芑蚧顒拥某晒?#xff1f; 在某種程度上，這些是分析人員必須做出的主觀決定才能產(chǎn)生任何結果。 但是同時，這些可能會對研究的實際結果產(chǎn)生重大影響。

UnsplashUnsplash

無意識的偏見 (Unconscious Biases)

Unconscious biases are not introduced by the analyst’s active decision making. This sort of bias is less connected to the personal opinion about a specific idea or research question but can have the same magnitude of impact on the results. Biases that fall into this category are for example:

分析人員的主動決策不會引入無意識的偏見。 這種偏見與對特定想法或研究問題的個人看法聯(lián)系較少，但對結果的影響程度相同。 屬于此類的偏差例如：

a) Programming Errors. Be it an error in a SQL-query, a wrong logic statement when filtering a pandas dataframe or an incorrect regex expression. All these programming errors can occur when we’re trying to get insights from the data in front of us. Other than syntax errors, this kind of programming error can remain wholly unnoticed when running our code and thus can have a substantial impact on the results of our analysis.

a)編程錯誤。 是SQL查詢中的錯誤，過濾熊貓數(shù)據(jù)框時的錯誤邏輯語句還是不正確的正則表達式。 當我們試圖從眼前的數(shù)據(jù)中獲取洞察力時，所有這些編程錯誤都可能發(fā)生。 除了語法錯誤外，在運行我們的代碼時，這種編程錯誤可能仍然完全未被注意到，從而可能對我們的分析結果產(chǎn)生重大影響。

b) Wrong handling of data. Usually, the data we want to examine to answer a particular research question does not come in a usable format. Before we can use a statistical model to derive insights from our data, we might have to clean it, select and engineer appropriate features, and eventually perform data transformations. All those actions can bias our dataset and thus our decisions in one direction or another.

b)錯誤處理數(shù)據(jù)。 通常，我們要檢查以回答特定研究問題的數(shù)據(jù)不是可用的格式。 在使用統(tǒng)計模型從數(shù)據(jù)中獲取見解之前，我們可能必須清理數(shù)據(jù)，選擇和設計適當?shù)墓δ?#xff0c;并最終執(zhí)行數(shù)據(jù)轉換。 所有這些動作都會使我們的數(shù)據(jù)集產(chǎn)生偏差，從而使我們的決策朝著一個方向或另一個方向傾斜。

c) Wrong interpretation of data. We might have done everything done when handling and modelling our dataset. But in the end, we can still derive the wrong conclusions from the results in front of us. Classic misinterpretations are confusing correlation with causation or drawing the wrong conclusions about the relationship of two parameters.

c)數(shù)據(jù)解釋錯誤。 在處理和建模數(shù)據(jù)集時，我們可能已經(jīng)完成了所有工作。 但是最后，我們仍然可以從我們面前的結果中得出錯誤的結論。 經(jīng)典的誤解使因果關系變得混亂，或者對兩個參數(shù)之間的關系得出了錯誤的結論。

那么分析師是不值得信任的嗎？ (So is the analyst to be mistrusted?)

Today’s most valuable companies such as Netflix, Amazon and Google show that experimentation and data have to replace intuition as a basis for making decisions.

如今，諸如Netflix，Amazon和Google之類的最有價值的公司表明，實驗和數(shù)據(jù)必須取代直覺作為決策的基礎。

Hence having trust in the data and therefore the analyst’s output is essential. It’s the analyst’s responsibility to build and maintain that trust. Analysts have to do their best to provide unbiased, informative insights to support decision-making and drive businesses in the right direction.

因此，對數(shù)據(jù)以及對分析人員的輸出的信任至關重要。 建立和維護這種信任是分析師的責任。 分析師必須盡力提供??公正，有用的見解，以支持決策制定并推動業(yè)務朝著正確的方向發(fā)展。

Therefore, it is imperative to be aware of your own biases and to overcome them where possible.

因此，必須意識到自己的偏見并在可能的情況下克服它們。

When being asked for a rough estimate or when thinking about how deep you want to drill into a specific topic, take a step back. Reflect on your decision and thought process and try to get a neutral perspective on your current issue.

當被要求進行粗略估算或考慮要深入到特定主題的深度時，請后退一步。 反思您的決策和思考過程，并嘗試對當前問題持中立觀點。

To avoid unconscious biases, it helps to adopt some best practices from the world of software engineering: Use unit tests in your queries and notebooks, start pair programming and ask colleagues to review your code and approach.

為避免無意識的偏見，它有助于采用一些軟件工程領域的最佳實踐：在查詢和筆記本中使用單元測試，開始結對編程，并要求同事審查您的代碼和方法。

Liked this article? Then you might also like one of my other posts:

喜歡這篇文章嗎？ 然后，您可能還會喜歡我的其他帖子之一：

翻譯自: https://towardsdatascience.com/the-analysts-bias-5c84825c0f48

標記偏見

總結

以上是生活随笔為你收集整理的标记偏见_分析师的偏见的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內容還不錯，歡迎將生活随笔推薦給好友。