Chapter 2 1.2 Types of Data Analysis
Before talking about R and social problems, let’s talk about the types of data analysis first.
Leek & Peng(2015) categorized data analysis into the 6 types as presented in the table below, and“mistaking the type of question being considered is the most common error in data.”
| Types of Data Analysis | Questions being asked |
|---|---|
| Descriptive data analysis(기술통계분석) | seek to summarize(요약) the measurementa single data set without further(e.g., US Census) |
| Exploratory data analysis(탐색적자료분석) | search for discoveries(발견), trends,, or relationships between theto generate ideas or(e.g., The four-star planetaryTatooine) |
| Inferential data analysis(추론분석) | quantify whether an observed pattern will likelybeyond the data set in hand or in**population**(모집단)(e.g., a study of whether aircorrelates with life expectancy in US) |
| Predictive data analysis(예측분석) | predict(예측) another measurement (the) on a single person or unit(e.g.,of how people will vote in an) |
| Causal data analysis(인과관계분석) | seek to find out what happens to oneon average if you make another*change**(인과, e.g., causalbetween smoking and cancer) |
| Mechanistic data analysis(결정론적관계분석) | seek to show that changing one measurementand exclusively leads to a specific,behavior in another(인과, e.g.,design) |
From Leek & Peng (2015)
Leek & Peng(2015)’s main point is that we should keep in mind the type of question being asked by our own data analysis. In other words, we should say what we can say, not what we want to say.
Leek & Peng(2015) presents a table showing common mistakes
| Real Question Type | Perceived Question Type | Phrase Describing Error |
|---|---|---|
| Inferential | Causal | Correlation does not imply causation |
| Exploratory | Inferential | Data Dredging (or p-hacking) |
| Exploratory | Predictive | Overfitting(과적합) |
| Descriptive | Inferential | n of 1 analysis |
- Leek & Peng(2015) mentioned a very important statement that we should keep in mind when we use data science to solve social problems:
“In nonrandomized experiments, it is usually only possible to determine the existence of a relationship between two measurements, but not the underlying mechanism or the reason for it.”
It is known that the best way to investigate causal relationship is to conduct randomized experiments. However, unlike in natural science, it is not easy to conduct randomized experiments in social science because of ethical and practical reasons. The fundamental dilemma of data analysis in social science is that we essentially want to make causal statements in the absence of randomized experiments. Many statistical tools we use are just correlational.
In the field of the philosophy of science, it is usually said that the goals of science is explanation and prediction:
“Historically, social scientists have sought out explanations of human and social phenomena that provide interpretable causal mechanisms, while often ignoring their predictive accuracy. We argue that the increasingly computational nature of social science is beginning to reverse this traditional bias against prediction.” (Hofman et al., 2017)