深入浅出数据分析

EDWARD TUFTE是数据图形化的最高权威，他的著作宛如一座奇妙的数据图形化博物馆，数据图形化有时被他称为认知艺术。数据分析的固定基本流程:确定---分解---评估---决策你对外界的假设和你确信的观点就是你的心智模型统计模型取决于心智模型，所以分析的重中之重是明确心智模型，最好使用正确的心智模型。心智模型应当包括你不了解的因素

印象最深的几点：1.假设检验那章的用因果网络做定量评估。通过已知因素排除最不可能的结果，再利用已知因素对剩余的各种结果作定量评价（五个减号到五个加号),求平均得最终结果。这种方法对我很有启发，可以对一些逻辑层面的东西作定量分析。2.对可能性定量。很可能，有可能，不可能，非常可能的可能性是：70%，50%，15%，90%等等，这又是一种定量分析的途径。3.贝叶斯概率，主观概率俄罗斯石油的案例确实生动地展示了主观概率的误导性，不能过分依赖直觉。4.直方图，散点图等简单的图表能直观地展示数据的整体概貌，为深入的分析指引方向。5.平衡预测的解释性和预测性（balance explanation and prediction)explanation 和 prediction两者不可兼得，预测的范围越广，预测的可靠性越低，预测的可靠性越高，预测的范围越小。6.正确的假设基于错误假设的分析会令人误入歧途。 Define=>Disassemble=>Evaluate=>Decide7.检验某个理论一定要有对照组，对照组的选择一定要控制各种非关注因素的干扰。8.启发式分析分析时去除不重要不易观测不易定量的因素，保留可观测的有说服力的因素。9.其他：统计学，Excel技巧，R，可视化，关系数据库，文本处理技能……数据分析之路，路漫漫其修远兮。

Chapter 1 Introduction to data analysis 1. Define by how much do you want to increase sales? how do you think we'll do it?/who should we focus on? how much of a sales increase do you think is feasible? are the target sales figures reasonable? market share? how are our competitors' sales? what's the deal with the ads and the social networking marketing budget? 2. Disassemble how do we increase sales? what do our best customers want from us? what promotions are most likely to work? how is our advertising doing? * your client's beliefs * your thoughts on the data * ask clients: * where would you say are the biggest gaps in your knowledge about …sales? * how confident are you that advertising has increased sales in the past? * who else might buy the product besides tween girls? 3. Evaluate 4. Decide 5. brief report: concise, professional, and direct! 1. context: your client's beliefs(refer 2.) 2. Interpretation of data: your thoughts on the data (refer 2.) 3. RecommendationChapter 2 Experiments * what could you do to start figuring our how to increase sales? * Interview the CEO to figure out how XXX works as a business * Do a survey of customers to find out what they're thinking * when analyzing surveys, pay attention to confounder * confounder: a difference among the people in your study other than the factor you're trying to compare that ends up making your results less sensible. * How to randomize? * in a spreadsheet, create a column called Random * first cell: =RAND() * same for remaining cells * sort by Random column

把大问题分为小问题