1: Data, Datasets, SPSS, and Frequency Distribution Tables

A. Welcome to statistics!

Statistics is useful for unveiling group patterns and variability. Statistics is an important gateway into understanding and describing the social world. Statistics can advance our knowledge, improve decision-making, and enable us to make statements about populations. Statistics are used in all kinds of literature, and by all kinds of people trying to influence you, so being able to evaluate and understand statistical claims being made is also a critical skill.

B. Unit of Analysis

What is the entity your data is about (what's each case represent)? What are you trying to make claims about? These answers need to match, and their answer is your unit of analysis.

When you look at a dataset, the unit of analysis is what each row (or case) represents—be it individual people, states, books, countries, songs, or something else. Often survey data is going to be of individual people, but even then the data may have been aggregated (grouped) such that you have data at a different level of analysis. For example, if you have graduation rates for different schools, the school is the unit of analysis.

While a study might give you ideas for future research at a different unit of analysis, you can only make claims at the unit of analysis you are using. For example, if I am studying university graduation rates (unit of analysis = individual universities), I may know that Harvard University has one of the highest graduation rates, around 98%. However, if I meet someone who goes to Harvard, I should not assume they will graduate, or if someone tells me they studied at Harvard, I should not assume they graduated (e.g., Bill Gates, Matt Damon, Mark Zuckerberg, William Randolph Hearst, and Pete Seeger all studied at Harvard but did not graduate). Likewise, I should not assume that the U.S. higher education graduation rate is comparable. (In reality, 64% of first-time, full-time undergraduate students who began seeking a bachelor's degree at a 4-year degree-granting institution in Fall 2014 had their degree within six years.)

C. Variables

Generally large social datasets will all have variables like age, gender, and race. Age is a variable because not everyone has to have the same age. If you are trying to ask a research question to find out if a relationship exists, you need to have variables that vary. If your variable is a constant, you won’t be able to find out relationships. For example, if you want to know if there is a relationship between age and political ideology, but all of the people in your dataset are the same age, you won’t be able to tell if political ideology varies as age changes. Variables can be all kinds of things, such as height, temperature, birth weight, favorite color, income, partisan affiliation, number of pets in household, state voter turnout, etc.

D. Data & Datasets

• It is estimated that less than 10% of U.S. drivers don't wear seatbelts, but that drivers without seatbelts constitute 50% of all vehicle passenger deaths.
• The average daily intake of plain water in the United States is about 4 cups per person.
• If current trends continue, within the next half-century Christians will constitute a minority of U.S.-Americans.

You have probably come across statistics that sound like these throughout your life. But where did they come from? These statistics were all the result of analyzing data, factual information that could be analyzed. The driver statistics came from the National Highway Traffic Safety Administration's (NHTSA) National Occupant Protection Use Survey and from Fatality Analysis Reporting System data NHTSA collects from every state government with information on all fatal crashes. The water statistic came from the National Health and Nutrition Examination Survey. The religion statistic was based on Pew Research Center analysis of their American Trends Panel, other Pew surveys, and NORC's General Social Survey. To do statistical analysis, first you need collected data. You can collect your own data through original research, or use data that has already been collected, and conduct secondary data analysis.

When I put a collection of data together in an organized fashion that can be used for analysis, it is called a dataset. In statistics, your dataset will be one file in spreadsheet form. A qualitative dataset could consist of something different like a set of interview transcripts.

Because we're dealing with a small population of movies, I could do my own mathematical calculations, for example to figure out what percent of the movies featured monarchies. However, you are limited in what analyses you can do by hand, and what happens when you have a dataset with hundreds or thousands of cases? Either way, I would rather make the computer do the math for me. It's easier and more likely to be accurate. There are a number of statistical software programs. In sociology, the most popular ones include: SPSS, Stata, SAS, and R. One advantage of R is that it is free to use. This textbook uses SPSS. Specifically, the instructions for SPSS and the outputs shared here use SPSS Version 29. Other versions will be similar but sometimes have small differences. I like SPSS because I think it has the easiest learning curve. Its interface looks very similar to other applications you might use, and you can use drop-down menus to run commands and/or use its programming language. While you need a license to use SPSS, as an alternative, you can do almost all the analyses in this textbook using an open-source statistical software called PSPP, which is very similar to SPSS. It’s available for download here.

When you use statistical software, you want a dataset where all of your data is represented by numbers, not labels like "refused to say" or "Disney." The numbers that represent certain characteristics, values, or categories are called codes.

When we enter data into a dataset to use in SPSS, we engage in coding, meaning we assign codes for each category of our variable. We then enter numbers into our dataset. For any given variable, a code matches particular numbers to particular responses. A codebook tells you the coding scheme for each variable—what the codes are for each variable and what they represent.

SPSS datasets are files that end in .sav (System for Analysis of Variables format). I can open up my Excel spreadsheet in SPSS as well, but all the data files you will be using will be SPSS files that end in .sav. To open up a dataset in SPSS, when you are in SPSS, go to File → Open → Data, and select your file.

SPSS datasets have a tab on the bottom that allows you to toggle between different views of the dataset. One view is called the Data View, which looks similar to the dataset above. Each column is a unique variable and each row is a unique case. The other view is called Variable View, which gives you information about the variables: their name, "label," which is a description of the variable that will appear in many outputs, "values," which is the coding for the variable, and "missing," which will specify if any codes are supposed to be excluded from analyses.

When there is no data for a particular case for a particular variable, in SPSS Data View you will see a . instead of a number in that particular cell. Answers can also be coded as "missing" if they are supposed to be excluded from your analysis.

Most of the time when you are told stories about data, you are not going to encounter the actual datasets. You are going to hear a story about aggregate, analyzed data. However, if you want to do your own data analysis, you need a dataset. We are learning social statistics, so are going to spend a lot of time with datasets. Datasets can come from data you collect yourself (e.g., your own original survey or content analysis), or you can engage in secondary data analysis—when you use existing data someone else already collected, but do your own analysis. There are plenty of public datasets available. Some with great social data include: the General Social Survey, U.S. Census Bureau datasets, the World Values Survey, Pew Research Center datasets, Uniform Crime Reporting Program data, the International Social Survey Programme, and American National Election Studies data. One popular source for finding data sets is the Interuniversity Consortium for Political and Social Research (ICPSR); there researchers can search over 3500 data sets to potentially find one that will address their research question.

In social statistics, frequency is synonymous with count. In some disciplines, the term frequency refers only to rates.

A frequency distribution table shows you each category of a variable and the number of cases for each category.

In addition to the count or frequency, frequency distribution tables have percentages and cumulative percentages. Percent means "out of 100." Cumulative is the total amount when you add everything together. So above, the cumulative percentage for Category B would be the total percentage for both Category A and Category B.

When you create a frequency table, you want to select a variable that does not have too many categories. Imagine if the Annenberg Report had included a frequency table with statistics for each of the 1,488 films! That would be a really long table. If you were making a frequency table broken down by age, and have a range of adult ages, you would likely want to collapse the age variable into age groups (e.g., Google Analytics does age reporting based on the groups 18-24, 25-34, 35-44, 45-54, 55-64, and 65+). If you were looking at voting preferences, you might combine any candidates with less than 1% support into an "other" category. There are times you may want a larger frequency table, such as if you are doing a state-by-state comparison and analysis. However, even then such a table is probably more useful to have and provide for reference and less useful for highlighting the main story of your findings.

F. Pie Charts

Pie charts are useful for showing relationships of parts to the whole when you do not have too many categories (slices of the pie).

G. Bar Graphs

Use bar graphs when you have frequencies of categories. If you use a bar graph on a variable with a Strongly Disagree to Strongly Agree scale, each bar will get a label stating what category it is. In Chapter 3 we will look at histograms. Those are a form of bar graph for when you have frequencies for numerical rather than categorical data (e.g., salary, number of pets, etc.).

H. Critical Consumption

I wouldn’t write a statistics book if I didn’t think statistics have awesome utility. But with a power tool comes responsibility. You should also be accurate and transparent when you share your own data analysis. You also have to be a critical consumer of statistics.