Scripps Employees Called 'Hackers' For Exposing Massive Security Flaw
Suspected PayPal Hackers Seek Deal To Stay Out Of Prison
Sophos polls again, claims 93% want opt-in sharing on Facebook
Their new poll asked whether Facebook should make sharing options opt-in or opt-out by default. An overwhelming majority of respondents -- 93% to be exact -- chose opt-in. That sounds huge, right?
Not so fast.
Here's the thing. This is based on a sampling of 605 people. Facebook's population has surpassed 400 million users, so this sample size really isn't sufficient. It's the equivalent of asking 500 Americans which is better, Coke or Pepsi, lucking into 450 who choose Pepsi, and then announcing "90% of Americans prefer Pepsi to Coke!" To be fair, Sophos does spell out that the poll is unscientific, but we'll see what the blogosphere does with the headline this time.
While I'd like to believe that so many Facebook users are clamoring for more control, the reality is much different. I'd wager that the majority of Facebook's 400+ million users probably don't even know what opt-in and opt-out mean. It would be nice if they did, and it would be nice if they cared -- but based on what I see on my wall every day, average users simply aren't that concerned. They're too busy playing Farmville and uploading pictures.
Comments
7
Subscribe to commentsMatt WaltonMay 27th 2010 11:24AM
The more data shared, the more new things for people on FB, the longer they will stay on the site/ the more frequently they will check it/ more pageloads. From a purely economic standpoint, it makes sense that they want people to share as much data as possible.
Andrew JaffeMay 27th 2010 11:45AM
You're probably right that this poll is skewed, but not for the reason you state. Polling only a small fraction of a large population is exactly the way polls work (but why you need to quote a "margin of error" which are roughly 2.5% for 1600 people and 5% for 600).
The real problem here is bias: most of the people likely to respond to the poll are those that really care about the issue, which are likely to be heavily weighted towards privacy advocates.
Lee MathewsMay 27th 2010 11:46AM
Yah, you're definitely right on that point, Andrew. While I did mention the general apathy of avg. Facebook users toward the end, I kind of neglected to tie that back to the fact that those people need to be polled for this to have _real_ value.
Bilal BarakatMay 27th 2010 11:49AM
You are mistaken. Actually the size of the sample you need for robust conclusions does NOT depend on the population size, if you have a truly random sample. The percentage of the total population polled is pretty much irrelevant (unless close to 100%). Think of it this way: you check one spoonful from a small saucepan for seasoning; do you need to check two spoons from a large saucepan? As long as it's mixed properly, you only need one spoonful regardless.
The problem with this poll is not the sample size, much less the share of the sample among all FB users, but the fact that it's not properly random.
Scott CookMay 27th 2010 11:54AM
No matter how you want to spin it you're not going to "luck" into getting 93% of the respondents to prefer "opt-in" and the next day"luck" into getting 93% to think the other way. Even if you are off by 25% it's still over 2/3 of the respondents preferring "opt-in". People want the option to make their own decisions not to have someone else decide for them (with an obvious bias that it's better for FB) and say well you can go and change your mind but you have to go through this convoluted process to change it.
There's been certainly enough backlash to indicate that opt-out is not what users want. What people continue to post would not change the answer to this question. If the question were "Do you post differently because you were not asked to opt-in?" the percentage would probably also be about 93% or higher in the No direction supporting your conclusion.
boo_radleyMay 27th 2010 12:46PM
It might be counterintuitive to you, but you really can get a good feel for large populations with small samples. I say that as an analyst who works with statistical data on a regular basis. Additionally, since the poll has only two choices, rather than a broad spectrum of possible responses, and it's entirely believable that the sample really does represent the whole. In fact, doing a few quick calculations shows that to sample 400,000,000 people at 95% confidence with a 5 pt. margin of error, you'd only need to sample around 380 people.
More relevant is the manner in which the poll is conducted: voluntary response polls tend to report initially to one side, but when an issues is politically volatile, the poll can often swing counter as the "other side" mobilizes users to vote.
michas_piMay 27th 2010 6:29PM
A sample size of thirty or more is sufficient to establish a correlation.