Understanding Data

Sociology 312/412/512, University of Oregon

Aaron Gullickson

What Does Data Look Like?

Most data looks like a spreadsheet

Four passengers on the Titanic

survival	sex	age	agegroup	pclass	fare	family
Survived	Female	24.0000	Adult	First	69.3000	0
Died	Male	24.0000	Adult	Third	7.7958	0
Survived	Male	0.9167	Child	First	151.5500	3
Died	Male	60.0000	Adult	First	26.5500	0

• Observations are on the rows. Observations are the units from which you are taking measurements.
• Variables are on the columns. Variables measure specific attributes of your observations.

The unit of analysis

The unit of analysis simply tells you what your observations are. We can collect data on many different types of units.

Some examples, but the possibilities are endless:

Individual People	Countries	Corporations	Universities

What is the unit of analysis?

1. Titanic data on passengers?
2. Cross-national data on CO2 emissions?

Titanic: Individual Passengers

Cross-national data: Countries

Quantitative variables

A quantitative variable measures quantities of something. A quantitative variable is always represented as a number. There are two types of quantitative variables.

Discrete

A discrete variable can only take certain values within a range.

• Number of children ever had
• Number of violent crimes committed
• Number of siblings
• Number of Youtube views

Continuous

A continuous variable can take any value within a given range.

• Age
• Height
• GDP per capita
• CO2 emissions

Categorical variables

Categorical variables indicate which category an observation belongs to from a mutually exclusive set of categories. There are also two types of categorical variables:

Ordinal

An ordinal variable is a categorical variables whose values have a clear ordering.

• Highest degree earned
• passenger class on the Titanic
• level of support with an opinion statement

Nominal

A nominal variable is a categorical variable whose values are unordered.

• Gender
• Race
• Political party

What variables types do we have?

Four passengers on the Titanic

survival	sex	age	agegroup	pclass	fare	family
Survived	Female	24.0000	Adult	First	69.3000	0
Died	Male	24.0000	Adult	Third	7.7958	0
Survived	Male	0.9167	Child	First	151.5500	3
Died	Male	60.0000	Adult	First	26.5500	0

• survival: survival is a nominal variable.
• sex: sex is a nominal variable.
• age: age is a continuous variable.
• agegroup: age group is an ordinal variable.

• pclass: Passenger class is an ordinal variable.
• fare: fare paid is a continuous variable (sort of).
• family: number of family members is a discrete variable.

What Can We Do With Data?

Look at the distribution of a variable

Look at association between variables

Make statistical inferences

• If I told you that in a sample of twenty people, brown-eyed individuals made $5 more an hour than people with other eye color, would you believe that I had found an important source of inequality?
• Probably not, because with a sample that small, its possible to get unusual results like this just because of the sample you happened to draw.
• Statistical inference is the technique of quantifying how uncertain we are about our statistical results. It allows us not only to summarize our data, but give a sense of how well we think that data might represent the larger population from which we drew our sample.

We can build models

In the last module of the course we will expand on what we have learned previously to build statistical models that can help us better understand relationships we observe in the data.

Statistical models allow us to examine relationships between multiple variables using more complex mathematical specifications. We use these models to address two types of questions:

• Do I observe a relationship between two variables, even after controlling for potential confounders?
• Does the relationship between two variables differ depending on the context of a third variable?

Models can get more complex

Confounders are variables that might account for the relationship between two other variables and thus produce a spurious non-causal relationship. Some real questions we might ask:

• Does the gender advantage in surviving the Titanic hold even after accounting for differences in gender and survival by passenger class?
• Do racial differences in the probability of being shot by the police hold even after accounting for situational factors such as the type of violation, time of day, place of interaction?
• Are differences in sexual frequency by marital status a product of age differences by marital status?

We can also go beyond simple statements about the association between two variables and begin to ask how the association between two variables might differ depending on the level or value of a third variable. Some real questions we might ask:

• Does the female advantage in survival on the Titanic get smaller or larger by the class of the passenger?
• Does the relationship between age and support for gay marriage differ by political party affiliation?
• Does the effect of movie runtime on box office returns vary by the genre of the movie?

Two views on observational data

Most of the data we use in the social sciences is observational rather than experimental. We observe what actually happens rather than manipulate “treatments” in order to observe a response.

There are two different views of exactly what the use of statistics contributes to observational data analysis:

Pseudo-Experimental

• We use statistical modeling to try to mimic the stronger causal claims of experimental research.
• This can be as simple as “controlling” for other variables to mimic random assignment of a treatment, to the use of “natural” experiments.

Formal Description

• We use statistical modeling to describe what we observe in the data in a formal, systematic, and replicable way.
• We show how the data are either consistent or inconsistent with certain views of a social process as derived from theory.