Replies

Any poll as long as it involves random selection and has rules of conduct set up before its taken is VALID within the framework of the rules and the mathematics of probability theory.

You do know they had to call more than 1600 people to get 160 responses, right? Pew revealed that rather universal problem yesterday. What that means is that all the polls are not statistically reliable if only for the replacement dilemma.

Let me explain that briefly. You want to take a random poll and you need at least 20 responses to get what is known as a 95% confidence level. That is, that as long as you keep taking samples, the average of all your samples is not going to very more than 5% no matter how many more samples you take that meet your standard for selection.

But you have 2 variables related to major party candidates (O and R) Then you have 3 kinds of respondents ~ (REP, DEM and IND)

The more factors you include in your poll, the more you have to be concerned with the problem of too few responses (less than 20) In your smallest cell ~ e.g. you could have hundreds of O/DEM, R/REP, but maybe only 19 O/IND ~ so you'd need to keep sampling to get that cell up to 20 before you could call a halt. Then every cell and element in your survey would have at least a 95% level of confidence ~ and folks would say 'Yah, verily, look, a 95% level of confidence'

Most of that is already well understood by most folks by the time they get to posting on a political commentary board. What is not well understood by most of those folks, or even by many professionals in the polling profession, or even very many statisticians, is that if your well designed survey misses a sample, you need to think hard about the next sample ~ you don't just drop the ones you missed from your final computations and move on by substituting the very next sample.

You must record a missed sample which, for most of them will be a missed phone call, or maybe somebody answered, but they weren't an adult, or they were an adult but they told you to stuff it.

When polling for more than one characteristic ~ e.g. Republicans or Democrats or Independents voting for Republicans or Democrats, you must understand that each one of your missed samples actually was someone voting or not voting, and for or against Republicans or Democrats, and that person may have been a Republican, a Democrat or an Independent, but you don't know!

Yet, there they were, at call 39 as selected by your random dialing generator, and they didn't answer, and now you need to substitute a new sample for that missed sample.

If you simply move on to the next one you may be up against hidden factors that are going to make more of "That kind" not respond on that next sample. For instance, Republicans may be tied up watching the Republican convention ~ that would mean more Democrats were available for the calls than Republicans.

Or, it's daytime hours and Republicans are at work, or it's early evening hours and Republican shopkeepers are still at their stores, or it's late evening hours and Democrats are down at the club, or it's early morning hours and Independents are dragging in after a hard night doing stuff (whatever stuff is that Independents do).

I worked with systems that were so specific the sample selection clerk was told to go to THAT specific employee and look at what piece of mail he or she had in the right hand, or the left hand!

You couldn't miss those and just go find some other employee and check the wrong hand ~ oh, no, you had to report it as a missed sample, and accept different instructions for the substitute sample ~ that might not even occur that same day in that same building!

It's not any differrent with political sampling. You start at 6pm and end at 7pm, you have to make up any substitutes under the same conditions of original selection ~ that is, same area codes, same geographical location, in the same time frame.

I suspect the political science and math majors who make up the student groups working for the university systems go through all of that ~ until it comes to them without thinking. At the same time those polls can only take place during the hours the students are available ~ and that's where they go wrong. Student hours are abnormal. Only steel discipline can keep those polls on target ~

Here we have a poll with only 160 responses ~ but it had 5 factors ~ that means you had to have at least 20 Independents reporting Romney and 20 reporting Obama, and 20 Democrats reporting Romney and 20 reporting OBama, and 20 Republicans reporting Romney and 20 reporting Obama. That tells you all your sample cells have a 95% level of confidence. There's at least 20 in all of them, which means you have 120 of your 160 responses serving as a fundamental base ~ and an additional 40 responses differing from each other ~ and the poll's conclusions are based on just those 40 responses.

Does anyone see anything wrong with 40 responses guiding a pollster's discussion?

When it comes to QC systems that might be quite a goodly number, but you're not looking for quality comparisons ~ just that something meets a minimum standard, or it doesn't.

40 responses, 5 characteristics and some sort of margin of error.

BWAhahahahahhahahah ~ does not compute folks. You might do this if you are doing a series of quickies over a period of time to find a trend under changing conditions, but it really can't tell you who is ahead of the other guy.

I suspect some college educated statistician has started up a tracking poll that depends on a series of small polls ~ with a relatively large margin of error, and his boss, less educated but certainly politically opportunistic, has pulled this one item out of a series and said 'Lo, and behold, Romney's losing bad' and had that sent over to the 'news hole' by the editor who is even less knowledgeable in what this is all about ~ all he cares about is 'Romney Loses'.