Man, I remember the scare stories in 1991 about how Samoan kids in SoCal were at risk for cholera because, statisically, they had fewer toilets per capita. Public outcry for subsidized toilets was fierce, a commission was formed at the cost of millions.
Now I think I know what was behind it all. 
I have stated several times that I have no objection to the short form.
I’m not saying they treat it with respect. In the narrow area that we have been discussing…data collected for research purposes…the reason the data is not associated with individuals is not necessarily because of ethics; it’s because there is no motivation for doing so.
I already stated that I am not involved in sampling. However, although I agree that there would be selection bias, I would need someone to convince me that it would make a difference. We are talking about such a huge sample that I doubt there could be much fluctuation that makes a difference.
Um, the size of the sample shouldn’t matter for selection bias, right? This isn’t an issue of statistical consistency, where higher sample sizes lead to better results.
Time to fight some ignorance.
Look, I find it entirely conceivable that certain users of marketing data don’t have access to names and addresses.
But I’m surprised that Sarahfina appears to believe that consumer survey data doesn’t feed into the junk mail industry.
Bank rate monitor covered this topic:
I take it Sarahfina does not work in direct marketing. The article gets better:
Emphasis added: Direct mail companies gather disparate information and compile dossiers on individuals, name and address included.
Dennis won a civil suit against the company in question. But there is no indication that the relevant decision makers, who passed on her information to prison inmates, spent any time in jail.
Summary: Laws matter. Though Direct Marketers would doubtlessly love to get their mitts on individual Census records, they are deterred by poor access and the threat of prison.
Marketing surveys don’t present such hurdles – with predictable results. Those who fear the government yet trust for-profit institutions might reflect upon their understanding of contemporary democratic market economies.
You don’t sound very sure.
As far as I know, enlarging the sample size will smooth out all types of bias. They also do what they call “weighting” data. In fact, I believe the Census Bureau already does this…for example, they weight the data for illegal immigrants, because they know that many of them do not comply with the census. This is standard practice in data collection.
I know for a fact they don’t, because, as an employee of a research firm, I have personally been responsible for the destruction of identifying information. It is never delivered to our corporate clients. Only the data in aggragate is.
This is getting silly. Warranty cards are not “consumer survey data.” I do not have any experience with direct marketing because I work in market research which is an entirely different and completely unrelated field.
That’s terrific.
I still don’t see any evidence that identifying information gathered in marketing surveys is sold to direct marketers.
ETA: BTW, my name is spelled “Sarahfeena.” If you are going to repeat it so many times, you might as well spell it right.
Apologies.
Er Sarahfeena: Warranty cards are surveys of customer opinions.
I’m glad your firm destroys identifying information. But not all firms do that. Furthermore, I don’t see how the consumer is suppose to reliably distinguish between firms that destroy IDs and those who don’t. Are there penalties for misuse of this information?
I’ve shown multiple examples of corporate violations of privacy. None have been shown for the census. Let’s rewind the tape:
Heh. Both direct mail and credit card databases are “Corporate collected data”. I guess you meant to say “Some types of corporate collected data, but my no means all types of corporate collected data, are never associated with the individuals it is generated from”. But that’s not much of an argument, is it?
Now frankly, I am not familiar with Sarafeena’s subindustry, so googling for scandals is difficult. But it really doesn’t matter does it? The point is that to trust corporate data collection procedures over the Census is to place faith over fact.
Well, my statement was absolutely correct in a multivariate regression context. And frankly, I’m a little embarrassed at what appears to be a display of false expertise. But I have a better background in estimation: my survey design knowledge is limited and indirect.
In reverse order, weights are typically derived from the Census. A stratified sample is usually a weighted sample.
And yes, weighting can correct for bias, if you have the proper weights. But merely increasing the sample size won’t do squat. Increasing the sample will increase the efficiency of the estimate --it will lower the standard errors-- but that’s a different issue.
It’s a code of market research ethics that identifying information is not turned over to client companies. I have no idea if there are legal penalties, because no company I have ever worked for would have thought of doing such a thing. I have no idea what the policies of other industries are.
Maybe not, but I was trying to compare the same type of information, gathered in the same way, as census information. I was not lumping in demographic data taken from warranty cards. Once again, I’d like to point out that ALL of this information is voluntarily given to the company by the customer. If they don’t want to give it, they don’t have to. When I said that I might be “more likely” to give a corporation information about me, which is where this turn of the conversation seems to have come from, it was to illustrate the point that my problem with the Census long form isn’t because I’m too paranoid to release the information, it’s that I would like to be able to choose whether or not I do so, and to whom. That’s all.
Again, I don’t necessarily “trust” corporations more than government. As I said, either one needs a motivation to use my information negatively…no one is going to do it just for the hell of it.
Ok. Though I might add that “Voluntary” has a somewhat narrow meaning in some cases.
Sure, I “Voluntarily” give information to credit bureaus when I apply for a credit card – I don’t have to own a credit card after all, nor is anybody forcing me at gunpoint to acquire a checking account.
For myself though, the inconvenience of shifting to wholly cash-only transactions certainly exceeds $100 every ten years, which is the potential penalty for not filling out a census form.* I feel greater external pressure to share my information with banks: of course a sense of civic responsibility does not apply in these private contexts.
FWIW,
I found a primer on survey design on the web:
They note explicitly that increasing the sample size will not reduce survey bias. The page is not rocket science, but it contains the sort of material that I’m reluctant to make blanket statements about, absent some study.
The difference is that a private company can’t compel me to give them jack.
Ooooh kaaay. Lets put some methods behind some of these arguments points of discussion.
The Census short form, as we all know, represents the attempt to enumerate every person living in the United States, as of a particular date. It is conducted once every ten years. We never get it quite right, and there are arguments about how not right is, but we don’t really care about that now.
The long form, which is now being replaced with the American Community Survey, is sent to a sample of households, designed to be representative of the population.
The sample size is determined by the need to be representative of the US population, across various dimensions of the population. I didn’t spend the time to look up exactly what they were selecting for. If you want to, feel free (it’s 423 pages, have a blast): http://www.census.gov/acs/www/Downloads/tp67.pdf
OK, statistical terms.
Weighting - some population groups are oversampled in order to avoid one kind of measurement error. For example, probably more Native Alaskan households are included in the sample than would be if the survey was just randomly mailed out. Then, when the study is analyzed, the weights are applied so that the study becomes once again representative of the population. (Weighting is hard, let me know if you want more info.)
Now we get to Bias. Bias is Bad. Selecting a study population without paying attention to it’s representativeness (if that’s a word) would be bad - because then you would be spending a lot of money to do a study that told you a bunch of stuff about those people only - and nothing about the rest of the people. THAT would be a waste of money. That’s called Selection Bias.
**
One kind of measurement error **(or Measurement Bias) would be if you just let people answer your questions if they felt like it, and made no attempt to follow up. (That’s one example, and I think it’s the major one in play, here.
On a regular study, say in a University, if we randomly (or quasi-randomly) pick your name for a study, we will hound you. We will call at all times of day and night until we reach you. When we do, we make the heart-wringing pleas for cooperation (cure cancer, end heart disease …) We do pretty good. Oh - and we do know your name.
Only the Government has the power to compel, and only for the Census. I can see where it pisses people off. But if you don’t like it for it’s intrusiveness, then maybe do it for the statisticians! We have a hard life!
There are a lot of other kinds of Bias. The same kinds of Bias apply in different ways. Knowing them is why they pay us the big bucks. :rolleyes:
Oh - and someone said something about the CDC and being your Nanny and why don’t I go back to curing cancer and heart disease? I’m too tired to go back and look it up.
Get a freaking clue. What to you think we’re doing? Do you think you know that eating high fat diets leads to heart disease and cancer because we sat around and talked about it? We know about it because of large government-funded studies, and you know about it because we fund programs to let you know. If you want to go out smoking cigars and eating bacon-cheeseburgers, fine. But don’t get all whiny in the hospital when they won’t do the quadruple bypass you need because your lungs are in such bad shape that you won’t survive the anesthesia.