> shinypenny wrote: (Posting this to a new thread, since Google hasn't caught up, and I figured I'd forget by morning.) Just double-checked and you were right and I was wrong. Minimum sample size for any population is at bare minimum 30 (even for a population of 50), and more depending on the confidence level you're seeking. This is just based on a cursory 5-second Google search, but the table on this page (which refers one to a statistics book for more detail) was interesting: http://www.panasia.org.sg/iirr/ikmanual/sample.htm Based on this, if your population is 50,000, you'd only need to sample 400 people to have a 95% confidence level. (The other 5% are probably like you, and won't be bothered to talk to annoying surveyors!). Four hundred, to me, is still a pretty small survey though! jen I'm confused. You said you need a sample of 30, and then you said you need to sample 400 people out of 50,000, which is just under 1%. (I think it's 0.8% just looking at the numbers, actually, but I don't have my calculator handy).

Jen - can you explain this? (see below)
shinypenny wrote: (Posting this to a new thread, since Google hasn't caught up, and I figured I'd forget by morning.) Just double-checked and you were right and I was wrong. Minimum sample size for any population is at bare minimum 30 (even for a population of 50), and more depending on the confidence level you're seeking. This is just based on a cursory 5-second Google search, but the table on this page (which refers one to a statistics book for more detail) was interesting: http://www.panasia.org.sg/iirr/ikmanual/sample.htm Based on this, if your population is 50,000, you'd only need to sample 400 people to have a 95% confidence level. (The other 5% are probably like you, and won't be bothered to talk to annoying surveyors!). Four hundred, to me, is still a pretty small survey though! jen

I'm confused. You said you need a sample of 30, and then you said you need
to sample 400 people out of 50,000, which is just under 1%. (I think it's
0.8% just looking at the numbers, actually, but I don't have my calculator
handy).

"Bill in Co." <ornery@earthlink.net> wrote in message news:<cKFtb.2065$sb4.425@newsread2.news.pas.earthlink.ne t>... Jen - can you explain this? (see below)

Not sure what your question is.

jen

"shinypenny" <shinypenny0001@yahoo.com> wrote in message
news:c8cb5319.0311160910.55c3a584@posting.google.c om... "Bill in Co." <ornery@earthlink.net> wrote in message
news:<cKFtb.2065$sb4.425@newsread2.news.pas.earthlink.ne t>... Jen - can you explain this? (see below) Not sure what your question is. jen

I think he is asking about the difference between the minimum and maximum
sample size required for a certain confidence levels for different
population sizes. Lots of people think it is linear - if you need a certain
sample size for a population of 50, then you would need to sample the same
*percentage* of the population if you were dealing with 50,000. It just
doesn't scale up that way, though, which confuses the heck out of a lot of
people. It isn't intuitive at all. I was taught an analogy that was
helpful, though. If you are heating a bowl of soup and want to see if it is
hot enough, you stir it and take out a spoonful of soup and taste it. Your
sample size is one spoonful. If you are heating several gallons of soup for
a crowd, you stir it and take out a spoonful of soup and taste it. Your
sample size is STILL one spoonful, and you are just as confident that your
results are accurate, even though your "population size" is much bigger.

OK, let me try again, with the post intact for reference, Jen: (pse see
below)

Bill in Co. wrote: Jen - can you explain this? (see below) shinypenny wrote:> (Posting this to a new thread, since Google hasn't caught up, and I> figured I'd forget by morning.)>> Just double-checked and you were right and I was wrong. Minimum sample> size for any population is at bare minimum 30 (even for a population> of 50), and more depending on the confidence level you're seeking.>> This is just based on a cursory 5-second Google search, but the table> on this page (which refers one to a statistics book for more detail)> was interesting:>> http://www.panasia.org.sg/iirr/ikmanual/sample.htm>> Based on this, if your population is 50,000, you'd only need to sample> 400 people to have a 95% confidence level. (The other 5% are probably> like you, and won't be bothered to talk to annoying surveyors!).>> Four hundred, to me, is still a pretty small survey though!>> jen

I'm still confused. I am trying to relate these two numbers, and I'm
obviously missing something here:

You said above that you need a bare minimum sample size of 30, and then you
said "based on this....." that you need to sample 400 people out of 50,000 to
achieve a confidence level of 95%, which is just under 1%. So which is the
correct answer is what I'm asking. Or are they both? You need a minimum
sample size of 30 in ANY case, AND, if you want a 95% confidence level, you
need a 0.8% sample size.

"Joy" <fairly_happy_doesn't_need_any_more_spam@withoutspa myahoo.com> writes:
"shinypenny" <shinypenny0001@yahoo.com> wrote in message news:c8cb5319.0311160910.55c3a584@posting.google.c om... "Bill in Co." <ornery@earthlink.net> wrote in message news:<cKFtb.2065$sb4.425@newsread2.news.pas.earthlink.ne t>... Jen - can you explain this? (see below) Not sure what your question is. jen I think he is asking about the difference between the minimum and maximum sample size required for a certain confidence levels for different population sizes. Lots of people think it is linear - if you need a certain sample size for a population of 50, then you would need to sample the same *percentage* of the population if you were dealing with 50,000. It just doesn't scale up that way, though, which confuses the heck out of a lot of people. It isn't intuitive at all. I was taught an analogy that was helpful, though. If you are heating a bowl of soup and want to see if it is hot enough, you stir it and take out a spoonful of soup and taste it. Your sample size is one spoonful. If you are heating several gallons of soup for a crowd, you stir it and take out a spoonful of soup and taste it. Your sample size is STILL one spoonful, and you are just as confident that your results are accurate, even though your "population size" is much bigger.

That's great analogy!

I'd say it is good in a non-obvious way too. For your sample to be
useful it needs to be random, which requires that you stir the soup
first.

It is a little harder to stir a large pot well than a small pot.
Similarly, it can be harder to get a random sample from a large
population than from a small one.

Doug Anderson wrote:
I was taught an analogy that was helpful, though. If you are heating a bowl of soup and want to see if it is hot enough, you stir it and take out a spoonful of soup and taste it. Your sample size is one spoonful. If you are heating several gallons of soup for a crowd, you stir it and take out a spoonful of soup and taste it. Your sample size is STILL one spoonful, and you are just as confident that your results are accurate, even though your "population size" is much bigger. That's great analogy! I'd say it is good in a non-obvious way too. For your sample to be useful it needs to be random, which requires that you stir the soup first. It is a little harder to stir a large pot well than a small pot. Similarly, it can be harder to get a random sample from a large population than from a small one.

And if your pot of soups is not a homogeneous mixture (lets say you have a pot of
colored marbles instead!), then the size of the sample matters, doesn't it?

Ellie <ellie_first@hotmail.com> writes:
Doug Anderson wrote: I was taught an analogy that was helpful, though. If you are heating a bowl of soup and want to see if it is hot enough, you stir it and take out a spoonful of soup and taste it. Your sample size is one spoonful. If you are heating several gallons of soup for a crowd, you stir it and take out a spoonful of soup and taste it. Your sample size is STILL one spoonful, and you are just as confident that your results are accurate, even though your "population size" is much bigger. That's great analogy! I'd say it is good in a non-obvious way too. For your sample to be useful it needs to be random, which requires that you stir the soup first. It is a little harder to stir a large pot well than a small pot. Similarly, it can be harder to get a random sample from a large population than from a small one. And if your pot of soups is not a homogeneous mixture (lets say you have a pot of colored marbles instead!), then the size of the sample matters, doesn't it?

Your pot of soud _isn't_ homogeneous. That is one of the brilliant
things about this example. But the marbles are so small that a
spoonful is an awfully big sample.

"Ellie" <ellie_first@hotmail.com> wrote in message
news:3FB7D418.ADFC00B7@hotmail.com... Doug Anderson wrote: I was taught an analogy that was helpful, though. If you are heating a bowl of soup and want to see if
it is hot enough, you stir it and take out a spoonful of soup and taste it.
Your sample size is one spoonful. If you are heating several gallons of
soup for a crowd, you stir it and take out a spoonful of soup and taste it.
Your sample size is STILL one spoonful, and you are just as confident that
your results are accurate, even though your "population size" is much
bigger. That's great analogy! I'd say it is good in a non-obvious way too. For your sample to be useful it needs to be random, which requires that you stir the soup first. It is a little harder to stir a large pot well than a small pot. Similarly, it can be harder to get a random sample from a large population than from a small one. And if your pot of soups is not a homogeneous mixture (lets say you have a
pot of colored marbles instead!), then the size of the sample matters, doesn't
it?

Well, in this case if you leave everything cooking long enough for all
ingredients to reach thermal equilibrium then they should all have the same
temperature....

Ellie <ellie_first@hotmail.com> wrote in message news:<3FB7D418.ADFC00B7@hotmail.com>... Doug Anderson wrote: I was taught an analogy that was helpful, though. If you are heating a bowl of soup and want to see if it is hot enough, you stir it and take out a spoonful of soup and taste it. Your sample size is one spoonful. If you are heating several gallons of soup for a crowd, you stir it and take out a spoonful of soup and taste it. Your sample size is STILL one spoonful, and you are just as confident that your results are accurate, even though your "population size" is much bigger. That's great analogy! I'd say it is good in a non-obvious way too. For your sample to be useful it needs to be random, which requires that you stir the soup first. It is a little harder to stir a large pot well than a small pot. Similarly, it can be harder to get a random sample from a large population than from a small one. And if your pot of soups is not a homogeneous mixture (lets say you have a pot of colored marbles instead!), then the size of the sample matters, doesn't it?

If the question one is trying to answer is the incidence of
mis-attributed paternity in the whole country, I think there would be
some formidable problems. There is every reason to think that the
incidence of this event would not be consistent across social class,
etnic, geographic, and age categories. Sampling would need to
widespread thru out these categories to be meaningful, regardless of
sample size. Also, privacy and other considerations dictate that this
will never happen.
The current state of things is that paternity teating is more or
less routine in some parts of the country where fathers can afford and
it is allowed before child support in a divorce situation is
finalized. However, there is no centralized data collection and
reporting going on. Many testing companies occasionally mention the
high numbers of oopsies they are seeing. So, yes, my claims and
comments do not pass a statistical challenge. On the other hand, the
basic message from the available results is clear enough.
1. The rate of surprises is surprising to most guys.
2. The groups of families sampled the most so far are not poor,
nothing-to-lose type of people.
3. If one assumes that what is behind this phenoenon is a female
instinct to seek out an different type of man for breeding encounters
than her workaday husband,then it does not follow that this kind of
thing happens more often in unhappy, divorcing families than stable
ones. When there are small children in the home, it is nearly always
the wife who instigates the divorce. But why should having a "special"
child this way make her want to leave?
4. We may assume that women married to low status men would have
greater reason to resort to this method. But these families are not
yet being tested as often as higher status divorcing families.

Put all this together and I do not see any reason to believe that
the shocking results found currently would be any lower if paternity
testing were comprehensive.

rdubose@pdq.net (Ralph DuBose) writes:
If the question one is trying to answer is the incidence of mis-attributed paternity in the whole country, I think there would be some formidable problems. There is every reason to think that the incidence of this event would not be consistent across social class, etnic, geographic, and age categories. Sampling would need to widespread thru out these categories to be meaningful, regardless of sample size. Also, privacy and other considerations dictate that this will never happen.

Yes. This would be a really difficult thing to actually measure or to
estimate in a statistically defensible way.
The current state of things is that paternity teating is more or less routine in some parts of the country where fathers can afford and it is allowed before child support in a divorce situation is finalized. However, there is no centralized data collection and reporting going on. Many testing companies occasionally mention the high numbers of oopsies they are seeing. So, yes, my claims and comments do not pass a statistical challenge. On the other hand, the basic message from the available results is clear enough. 1. The rate of surprises is surprising to most guys.

OK. Though I'd still like to see figures telling us what this rate
_is_. You've claimed 30%, but where does that come from? Presumably
_someone_ measured _something_ to get that number (well, or someone
made it up).
2. The groups of families sampled the most so far are not poor, nothing-to-lose type of people.

OK. But they _are_ people getting divorced where the man is disputing
child support. Hence that is the group they represent, and no other.
3. If one assumes that what is behind this phenoenon is a female instinct to seek out an different type of man for breeding encounters than her workaday husband,

Why would one make such an assumption?
then it does not follow that this kind of thing happens more often in unhappy, divorcing families than stable ones. When there are small children in the home, it is nearly always the wife who instigates the divorce. But why should having a "special" child this way make her want to leave?

In other words, if you _assume_ that being unfaithful is a basic
female drive which is frequently acted on then you conclude that being
unfaithful is a basic female drive which is frequently acted on.

It's pretty hard to argue with that.
4. We may assume that women married to low status men would have greater reason to resort to this method. But these families are not yet being tested as often as higher status divorcing families. Put all this together and I do not see any reason to believe that the shocking results found currently would be any lower if paternity testing were comprehensive.

Whoa.

In other words, you make an assumption about the nature of women which
allows you to generalize from the sample of "divorcing families where
child support is contested" to "all families."

It is an improvement from the logical point of view over claiming that
"divorcing families where child support is contested" is a good random
sample.

Now you've made it clear that your argument just involves basic
misogyny.

Ralph DuBose wrote:
So, yes, my claims and comments do not pass a statistical challenge. On the other hand, the basic message from the available results is clear enough. 1. The rate of surprises is surprising to most guys. 2. The groups of families sampled the most so far are not poor, nothing-to-lose type of people. 3. If one assumes that what is behind this phenoenon is a female instinct to seek out an different type of man for breeding encounters than her workaday husband,then it does not follow that this kind of thing happens more often in unhappy, divorcing families than stable ones. When there are small children in the home, it is nearly always the wife who instigates the divorce. But why should having a "special" child this way make her want to leave? 4. We may assume that women married to low status men would have greater reason to resort to this method. But these families are not yet being tested as often as higher status divorcing families. Put all this together and I do not see any reason to believe that the shocking results found currently would be any lower if paternity testing were comprehensive.

In other words first fix the result that you are seeking, then make a number of off the wall
assumptions to support that result. I like your method, you can't ever go wrong. We can call
this designer statistics!

Doug Anderson <ethelthelog@yahoo.com> wrote: Your pot of soud _isn't_ homogeneous. That is one of the brilliant things about this example. But the marbles are so small that a spoonful is an awfully big sample.

Depends on the soup; broth, plain tomato soup or cheese sauce are all
effectively homogenous, while in some particularly chunky cases a single
spoonful may not be enough.

trajan@sfchat.org (Marcus Ulpius Traianus) writes:
Doug Anderson <ethelthelog@yahoo.com> wrote: Your pot of soud _isn't_ homogeneous. That is one of the brilliant things about this example. But the marbles are so small that a spoonful is an awfully big sample. Depends on the soup; broth, plain tomato soup or cheese sauce are all effectively homogenous, while in some particularly chunky cases a single spoonful may not be enough.

Nah. They are all made out of particles, just differently sized ones.

You say they are "effectively homogeneous." But that is code for what
_I_ said; which is the samples are really really large.

Doug Anderson <ethelthelog@yahoo.com> wrote in message news:<a1Vtb.216175$HS4.1870692@attbi_s01>... rdubose@pdq.net (Ralph DuBose) writes: If the question one is trying to answer is the incidence of mis-attributed paternity in the whole country, I think there would be some formidable problems. There is every reason to think that the incidence of this event would not be consistent across social class, etnic, geographic, and age categories. Sampling would need to widespread thru out these categories to be meaningful, regardless of sample size. Also, privacy and other considerations dictate that this will never happen. Yes. This would be a really difficult thing to actually measure or to estimate in a statistically defensible way. The current state of things is that paternity teating is more or less routine in some parts of the country where fathers can afford and it is allowed before child support in a divorce situation is finalized. However, there is no centralized data collection and reporting going on. Many testing companies occasionally mention the high numbers of oopsies they are seeing. So, yes, my claims and comments do not pass a statistical challenge. On the other hand, the basic message from the available results is clear enough. 1. The rate of surprises is surprising to most guys. OK. Though I'd still like to see figures telling us what this rate _is_. You've claimed 30%, but where does that come from? Presumably _someone_ measured _something_ to get that number (well, or someone made it up).

The 30% figure comes from testing companies currently doing this
service. Of course, this is not a valid "sample" of the entire
population but for reasons listed, it is interesting even so.



2. The groups of families sampled the most so far are not poor, nothing-to-lose type of people. OK. But they _are_ people getting divorced where the man is disputing child support. Hence that is the group they represent, and no other.

Here is a point that is not getting thru. Dads in divorce cases are
not asking for testing just because they have had reason, in their own
minds, to suspect something is amiss. Testing is being done because
lawyers strongly recommend it. And how could it be otherwise with
these sorts of results?
3. If one assumes that what is behind this phenoenon is a female instinct to seek out an different type of man for breeding encounters than her workaday husband, Why would one make such an assumption?

Because they got pregnant and chose to bear the child when they
could have born her husbands child.



then it does not follow that this kind of thing happens more often in unhappy, divorcing families than stable ones. When there are small children in the home, it is nearly always the wife who instigates the divorce. But why should having a "special" child this way make her want to leave? In other words, if you _assume_ that being unfaithful is a basic female drive which is frequently acted on then you conclude that being unfaithful is a basic female drive which is frequently acted on. It's pretty hard to argue with that.


I was not making tht argument. I was merely pointing out that there
are few divorces resulting from a man suspecting the true paternity of
his kids because men in general rarely initiate divorce for any reason
when kids are small. So child support age kids, the ones being tested,
are not coming out of situations where the divorce is happening
because of a mans suspicion, it seems to me.


4. We may assume that women married to low status men would have greater reason to resort to this method. But these families are not yet being tested as often as higher status divorcing families. Put all this together and I do not see any reason to believe that the shocking results found currently would be any lower if paternity testing were comprehensive. Whoa. In other words, you make an assumption about the nature of women which allows you to generalize from the sample of "divorcing families where child support is contested" to "all families." It is an improvement from the logical point of view over claiming that "divorcing families where child support is contested" is a good random sample. Now you've made it clear that your argument just involves basic misogyny.


A very strange comment. Do women have any responsibility, in your
mind, for the reality behind these numbers?

rdubose@pdq.net (Ralph DuBose) writes:
Doug Anderson <ethelthelog@yahoo.com> wrote in message news:<a1Vtb.216175$HS4.1870692@attbi_s01>... rdubose@pdq.net (Ralph DuBose) writes: If the question one is trying to answer is the incidence of mis-attributed paternity in the whole country, I think there would be some formidable problems. There is every reason to think that the incidence of this event would not be consistent across social class, etnic, geographic, and age categories. Sampling would need to widespread thru out these categories to be meaningful, regardless of sample size. Also, privacy and other considerations dictate that this will never happen. Yes. This would be a really difficult thing to actually measure or to estimate in a statistically defensible way. The current state of things is that paternity teating is more or less routine in some parts of the country where fathers can afford and it is allowed before child support in a divorce situation is finalized. However, there is no centralized data collection and reporting going on. Many testing companies occasionally mention the high numbers of oopsies they are seeing. So, yes, my claims and comments do not pass a statistical challenge. On the other hand, the basic message from the available results is clear enough. 1. The rate of surprises is surprising to most guys. OK. Though I'd still like to see figures telling us what this rate _is_. You've claimed 30%, but where does that come from? Presumably _someone_ measured _something_ to get that number (well, or someone made it up). The 30% figure comes from testing companies currently doing this service. Of course, this is not a valid "sample" of the entire population but for reasons listed, it is interesting even so.

Can you give a single reference, on line or off? Or are you just
going to continue to assert this?
2. The groups of families sampled the most so far are not poor, nothing-to-lose type of people. OK. But they _are_ people getting divorced where the man is disputing child support. Hence that is the group they represent, and no other. Here is a point that is not getting thru. Dads in divorce cases are not asking for testing just because they have had reason, in their own minds, to suspect something is amiss. Testing is being done because lawyers strongly recommend it. And how could it be otherwise with these sorts of results?

Again, even if that were true, you don't have a representative sample.
Is that true? Is there something you can tell me beyond your
assertion?
3. If one assumes that what is behind this phenoenon is a female instinct to seek out an different type of man for breeding encounters than her workaday husband, Why would one make such an assumption? Because they got pregnant and chose to bear the child when they could have born her husbands child. then it does not follow that this kind of thing happens more often in unhappy, divorcing families than stable ones. When there are small children in the home, it is nearly always the wife who instigates the divorce. But why should having a "special" child this way make her want to leave? In other words, if you _assume_ that being unfaithful is a basic female drive which is frequently acted on then you conclude that being unfaithful is a basic female drive which is frequently acted on. It's pretty hard to argue with that. I was not making tht argument. I was merely pointing out that there are few divorces resulting from a man suspecting the true paternity of his kids because men in general rarely initiate divorce for any reason when kids are small. So child support age kids, the ones being tested, are not coming out of situations where the divorce is happening because of a mans suspicion, it seems to me. 4. We may assume that women married to low status men would have greater reason to resort to this method. But these families are not yet being tested as often as higher status divorcing families. Put all this together and I do not see any reason to believe that the shocking results found currently would be any lower if paternity testing were comprehensive. Whoa. In other words, you make an assumption about the nature of women which allows you to generalize from the sample of "divorcing families where child support is contested" to "all families." It is an improvement from the logical point of view over claiming that "divorcing families where child support is contested" is a good random sample. Now you've made it clear that your argument just involves basic misogyny. A very strange comment. Do women have any responsibility, in your mind, for the reality behind these numbers?

I'm not sure if these numbers are anything other than a phantasm in
your mind. But of course the individuals involved are responsible for
what they did. I don't see a way to make "women" collectively
responsible for the actions of certain individuals.

Joy wrote: "Ellie" <ellie_first@hotmail.com> wrote in message news:3FB7D418.ADFC00B7@hotmail.com... Doug Anderson wrote:> I was taught an analogy that was> helpful, though. If you are heating a bowl of soup and want to see if it> is hot enough, you stir it and take out a spoonful of soup and taste it.> Your sample size is one spoonful. If you are heating several gallons of> soup for a crowd, you stir it and take out a spoonful of soup and taste> it. Your sample size is STILL one spoonful, and you are just as confident> that your results are accurate, even though your "population size" is much> bigger. That's great analogy! I'd say it is good in a non-obvious way too. For your sample to be useful it needs to be random, which requires that you stir the soup first. It is a little harder to stir a large pot well than a small pot. Similarly, it can be harder to get a random sample from a large population than from a small one. And if your pot of soups is not a homogeneous mixture (lets say you have a pot of colored marbles instead!), then the size of the sample matters, doesn't it? Well, in this case if you leave everything cooking long enough for all ingredients to reach thermal equilibrium then they should all have the same temperature....

So what was the consensus in here (in regards to my original question)?
That you need at least 30 samples (at a minimum) AND you need 0.8% for a 95%
confidence level? Was that it?