1 of 23

Data for advocacy

Getting started:

Florence Nightingale, 1858

2 of 23

I found a number!

“

”

950,000 Americans suffer from heart disease.

3 of 23

?

What questions should you be asking

about this number

“

”

950,000 Americans suffer from heart disease.

What questions should you ask yourself about this number before using it to make an argument? (students respond and discuss)

Basic questions that should come up:

What is the source of this number? Can we track down the original source? (News media and social media generally are NOT the original source of statistics – statistics are usually derived from some kind of study, usually by academics or government agencies.)
Who carried out the research? What are their credentials and agendas, and how might those influence the results? What was their methodology? Was the study peer-reviewed? Was it published?
What kinds of biases and errors might be expected in this number?
How are they defining “heart disease”? For that matter, how are they defining “American”? Did they include people living in Puerto Rico and other U.S. territories? Did they include the undocumented population? Both populations are in the millions.

4 of 23

?

that this is the

actual number

What are the chances

“

”

950,000 Americans suffer from heart disease.

5 of 23

?

of a statistic

depend on

What does the accuracy

“

”

950,000 Americans suffer from heart disease.

If a statistic is an extrapolation from a subset to a whole, what does the accuracy of a statistic depend on? (students respond and discuss)

The accuracy of a statistic depends on a number of things, including choosing the right kind of statistical analysis for the situation, and sample size, but the single most important factor is usually how representative the sample is.

Interestingly, because of something called the central limit theorem, you don’t actually need a huge sample size in order to derive accurate estimates. A sample size of 30 will usually suffice, regardless of the size of the overall population, IF the sample is truly representative of the whole and not biased in any way. The best way we know of to generate a representative sample is to sample randomly, but even random samples have some chance of being biased. (For example, even if you poll a sample of randomly generated phone numbers, there is a small chance that every person you call will belong to the same political party.)

6 of 23

?

What can we do with this number

in our advocacy

“

”

950,000 Americans suffer from heart disease.

7 of 23

are good for showing the

size of the problem

Single numbers

“

”

950,000 Americans suffer from heart disease.

8 of 23

single

numbers

two

numbers

(or a few)

a lot of

numbers

(a dataset)

sizes

comparisons

patterns

0

1

2

3

4

5

9 of 23

are the result of analyzing datasets

Single statistics

single

numbers

two

numbers

(or a few)

a lot of

numbers

(a dataset)

10 of 23

are the result of analyzing datasets

Averages

The median home price in Boulder, Colorado is $877,500

US life expectancy: 77.28 years

The mean age of the US population is 38.8 years

The average American makes $31,113 annually

11 of 23

68%

12%

22.3%

50%

0.05%

are the result of analyzing datasets

Percentages

12 of 23

are the result of analyzing datasets

Visualizations

13 of 23

usually have rows and columns

Datasets

Anyone who wants to engage in data advocacy needs to be able to tell the difference between datasets and statistics derived from datasets. Datasets are more powerful because you can do your own analyses. You can also combine them with other datasets to make new analyses possible.

Most of the time, you can recognize datasets by the row-and-column formatting. But data summaries can come in rows and columns too. Think about whether you are looking at the most original form of the data, and the most fine-grained. For example, if the dataset was created by sending out surveys to individual people, can you see individual people’s answers in the dataset? If not, there is a more original form of it somewhere. But because of privacy and other concerns, the original form of the data isn’t always publicly available.

14 of 23

can also be archival

Datasets

15 of 23

can also be archival

Datasets

16 of 23

can take other forms

Datasets

17 of 23

Is it a dataset?

So now let’s practice your ability to tell the difference between a dataset and the results of analyzing a dataset. Is this a dataset? (students respond and discuss)

Answers:

the visualizations at upper and lower left are not datasets. A visualization generally doesn’t count as a dataset – visualizations are generated from datasets.
At upper right are a couple of median prices which also must be generated from a dataset (and thus are not the dataset).
At lower right, we do see something that looks like a dataset with rows and columns, and these numbers could be used as a dataset for certain purposes. But it’s also clear that there must be a finer-grained dataset behind it, because each row summarizes hundreds of home sales across an entire calendar year. Depending on your purpose, you might want to try to track down some of that finer-grained data instead of working what what you see here.

18 of 23

Is it a dataset?

19 of 23

Is it a dataset?

20 of 23

Is it a dataset?

21 of 23

a dataset?

download data

scrape data from the web

generate your own data

How do I get

(e.g. by writing a script)

(e.g. by conducting a survey)

request data

If you want to find a dataset for use in advocacy, you have four basic options, listed here from easiest to hardest.

Downloading data is the simplest option, if somebody has made the data available online. Government agency websites are great for this. There are also lots of online repositories of data like Kaggle.com. Most of these datasets can be found via savvy internet searches.

Requesting data is the next easiest option. If you know about a dataset but it isn’t publicly available online, you might be able to get it by requesting it from the original researchers. This is particularly likely to work if they are academics. However, there are no guarantees, and in general this requires more time than a simple download.

A lot of data is available online already, but not in an easy-to-analyze form. For example, the University of Colorado publishes an anonymized dataset of all its employee salaries online. However, they publish it via a Javascript application that only displays 20 or so rows of the dataset at one time. This makes it difficult to perform any kind of analysis on the data. One way to collect the data is to manually scroll through hundreds of pages and copy and paste the rows of data from each page. In the past, students have written scripts in Python to automate this process, thereby generating a usable dataset in much shorter time. Two things to consider if you want to attempt something similar:

Be sure you are not violating the terms of service of the app or website whose data you want to scrape. Most social media platforms and many other websites do not allow any kind of data scraping.
Keep in mind that large institutions often publish their data in hard-to-analyze formats on purpose, because they want to maintain the appearance of data transparency without actually making their data easy to analyze. This is because they want to stay in control of their own data-driven narratives.

The most labor-intensive way to get data is to collect it yourself. I generally discourage my students from pursuing this route because it takes a LOT of time, bureaucratic approval, and expertise to pull off. Even administering a simple survey requires expertise in survey design, often IRB approval, and some way of distributing the survey and driving traffic to it, all of which are heavy lifts under the best of circumstances.

22 of 23

data privacy laws

HIPAA

FERPA

Keep in mind:

Health Insurance Portability and Accountability Act

Family Educational Rights and Privacy Act

23 of 23

data privacy laws

HIPAA

FERPA

Keep in mind:

makes medical records hard to access

(e.g. drug use, sexual assault, mental health)

makes student educational records

hard to access (grades etc.)