Listen to the podcast:
{podcast id=14}
Chat
Computers and privacyNSA culling data and how much it can tell you
- The story Adam mentions comes from a Stanford study that looks at what you might be able to learn about people using the NSA’s metadata usage pattern.
- Since recording the show, David went and installed the app to participate in the study. It was private, in that Stanford doesn’t share your information with anyone, and only took a few minutes. The app is for Android only, but it’s interesting to take part in it.
- From David’s record, anyone could tell you where he does his shopping, his laundry, where he works, that he took a trip to Baltimore two months ago…
- And this is the On the Media story .
Going from street photos to SSN
- The researcher is Alessandro Acquisti, who is known for doing evocative research. In this case, you can quickly read the summary here.
- Here’s the David Bronner interview on WTF
- Adam has a New Season’s story to go with this, where he destroys a man’s faith.
Articles
Research methods: Validity and Reliability
Every experiment has a wide opening and a narrow base.
ValidityHere's a thought experiment:
| Research question: | How much does jogging contribute to a person’s health? |
| Subjects: | 40 immobile computer scientists. Randomly distributed into two groups. |
| Group A: | 20 subjects paid to jog 5 miles per day for one month |
| Group B: | 20 subjects paid to not jog for one month |
| Independent variable: | (The thing we change) Jogging vs. not jogging |
| Dependent variable: | (The thing we measure) Body weight, before and after the study |
And we have some results:
Result: Group A loses about 15 pounds on average, Group B stays loses none on average
So it looks like jogging is conducive to good health.
But the important question: How do we define health?
Validity assessment: Bathroom scales measure only one thing: mass. They don’t tell you what kind of mass it is, just how heavy it is. But we talk about our weight, that is, the readout from the scale, as a representation of something else. That can be how much we eat, our general health, maybe even our height. The thing is, the scale is not a good tool for any of those things. It does not have validity for anything except mass. What more could we add to our dependent variable class that would improve our construct of health?
ReliabilityThe sensitivity of the scale is not the same as its accuracy. Say we have people get on and off the scale four times for their starting weight. Maybe we have something like this:
This table shows us a subset of some of the different weights that we recorded from the first encounter with the subjects. Sometimes the difference is minimal (e.g., participants 5 and 23), and sometimes it’s relatively large (e.g., participants 3 and 25). What do you think of the scale’s reliability?
Sound Science
Manipulating the direction of sound waves
The popular news article
And here's the journal article
David randomly mentions 3Dprinting of sand-grain-sized artifacts .
And in another aside, David mentioned futuristic(?) tech in the form of
Sound from Ultrasound
.
And yes, they've tried that.
TED Push
Lies, damned lies, and statistics about TEDtalks
.
And here is the
template for making the best (or worst) TED talk
ever.
Next time!
- A secret guest!
- Summarize the stats? Getting closer!
-----
(Cover image credit: http://emptysamurai.deviantart.com/art/Snake-oil-ad-289929870 )