The Top 1% is Stalling the Economy

[cybercoma]

No. It is the level at which the researcher *assumes* that it is not random. Even at that level there is still a 5% chance that it is actually random even if the researcher assumes otherwise.

But these 5% or 10% or 1% probabilities only apply if the researcher does not introduce selection bias by mining data for variables that happen to correlate because the researcher is can easily discarding the 95% of the data that does not correlate and only keep the 5% that does correlate. This is why I say the chances of random correlation goes up significantly when data mining goes on. This chance can be reduced if there is some independent evidence that the correlation should exist but in this case there is no independent evidence - all of the evidence is the correlation.

Wrong. At a basic level all data is sampled (i.e. a discrete series of measurements over time). If you want to a physical measure like tree rings and use it to predict temperatures then you use p-levels to show that the relationship between your available tree ring data and temperature is not random. However, researchers in this field also introduce selection bias by discarding samples that don't correlate which renders their claims meaningless. But that does not stop them from making strong claims.

It is very easy to fool yourself with stats. You appear to like being fooled when it supports your pre-determined ideas.

I'm sorry, but it's obvious you know just enough to not know what the hell you're talking about.

The significance level and p-value are two different things. The significance level is chosen by the researcher when determining whether or not to assume that a given p-value is random chance. This is done when you take a sample and are making inferences about a larger population. Sticking with your example of tree rings, you can't measure every single tree on the planet, so you take a sample of trees. When the calculations are made an actual p-valie is produced that is either greater than or less than the significance level. The p-value is a discrete value that could be greater than or less than your significance level. If I choose 5% as my significance level that doesn't mean all of my results have a 5% chance of having random error (this is what your posts are suggesting). It means that the researcher will accept any p-value of .05 or less as not being from random chance. The actual p-value could be much less than .05, even if the significance level chosen but he researcher is that. The actual p-value could be less than .0001. You don't know unless it's reported. Moreover, good researchers will choose stricter p-values with larger data sets. In a set with 30,000 cases almost everything is significant because of the size of the sample, but I digress.

The point is that significance levels and p-values are used when taking a sample and making inferences about the population. In your example, you take a sample of tree rings, as opposed to measuring ALL of the tree rings. Bringing it back to the current discussion, we're not taking a sample of GINI coefficients and GDPs from the OECD. We have ALL of the GINI coefficients and GDPs from the OECD nations. We're observing the entire target population of our study and not taking a sample. This is why significance levels and p-values are irrelevant here. We're directly observing the relationship between GDP or GINI with the social and health indicators.

It's quite obvious when you look at the charts that there is a relationship between GINI and the indicators, since they line up almost perfectly. You can see that the relationship between GDP and these metrics is not as strong because when you graph them the points are scattered. I feel I've explained this as far as I can with you because you're insistent on taking nonsensical diversions into methodology that you appear to have only a loose grasp on and that's being generous. You don't even have to understand statistical analysis to look at those two graphs and see which one (GDP or GINI) has a stronger relationship with the metrics. Your insistent denial is, as another member likes to say, "complete nonsense."

[TimG]

The significance level and p-value are two different things.

http://en.wikipedia.org/wiki/P-value

The p-value should not be confused with the Type I error rate [false positive rate] α in the Neyman–Pearson approach. Although α is also called a "significance level" and is often 0.05, these two "significance levels" have different meanings. Their parent approaches are incompatible, and the numbers p and α cannot meaningfully be compared. Fundamentally, the p-value does not in itself support reasoning about the probabilities of hypotheses, nor choosing between different hypotheses–it is simply a measure of how likely the data (or a more "extreme" version of it) were to have occurred by chance, assuming the null hypothesis is true

Translation: the p-value tells you the chance that the data is random even if the researcher assumes it is not.

It's quite obvious when you look at the charts that there is a relationship between GINI and the indicators, since they line up almost perfectly.

Who cares? Without some theoretical basis for such a relationship the lining up perfectly is meaningless.

If we were talking about a single statistic it would be more credible - the problem with these GINI studies is they selected a subset of the available statistics based on how well they correlate. That methodology is almost guaranteed to produce spurious correlations.

Edited January 8, 2014 by TimG

[cybercoma]

Show it. Show me where "they selected a subset of the available statistics based on how well they correlate." The stats they used to compare GDP and GINI were the major UN statistics for nations.

I hope while you do that you know that's the entire point of the discussion we've been having. To see whether there's a stronger relationship with GINI or GDP for certain social and health indicators. That there's other things that have a stronger relationship with GDP is nice and all, but show it. Wilkinson showed numerous indicators that have a strong connection to GINI, also showing that the aggregate of indicators (which aren't selected willy-nilly; they're actually the major UN stats) has a stronger relationship to GINI than GDP.

If you think the study is flawed and that the things they said, like infant mortality and homicide, have a stronger connection to GDP in wealthy nations, go ahead and show it.

Edited January 8, 2014 by cybercoma

[TimG]

Show it. Show me where "they selected a subset of the available statistics based on how well they correlate." The stats they used to compare GDP and GINI were the major UN statistics for nations.

If they really did that then there should be some stats which don't correlate. The fact that none of the stats the presented failed to correlate suggests that they selected the stats for correlation. If you want to claim otherwise then show examples of social metrics which fail the correlation test because statistical theory says there must be some even if the correlation is real (i.e. calculate the p-value for the collection of stats since they are making claims about the collection of stats and GINI). Edited January 8, 2014 by TimG

[Michael Hardner]

Seriously ? Now this is mostly an American issue ? Did you expect or prefer "global dominance" by Canada ?

Some people really don't know how to take a compliment.

[bush_cheney2004]

Some people really don't know how to take a compliment.

Welcome to the Top 1%.....anybody earning about $34,000 USD or more per year is part of this "elite" group. How can they be so selfish and greedy !

That was the finding World Bank economist Branko Milanovic presented in his 2010 book The Haves and the Have-Nots. Going down the distribution ladder may be just as surprising. To be in the top half of the globe, you need to earn just $1,225 a year. For the top 20%, it's $5,000 per year. Enter the top 10% with $12,000 a year. To be included in the top 0.1% requires an annual income of $70,000.

[Michael Hardner]

Welcome to the Top 1%.....anybody earning about $34,000 USD or more per year is part of this "elite" group.

Thanks... I... Huh ? How did I end up in THIS argument ?

This is like one of those old comedy movies where there's a square dance and through the comical changing of partners, the hero ends up dancing with the villain !

[cybercoma]

If they really did that then there should be some stats which don't correlate. The fact that none of the stats the presented failed to correlate suggests that they selected the stats for correlation. If you want to claim otherwise then show examples of social metrics which fail the correlation test because statistical theory says there must be some even if the correlation is real (i.e. calculate the p-value for the collection of stats since they are making claims about the collection of stats and GINI).

We're not talking about things that DON'T correlate with GINI. I'm pointing out to you that many things do. Now you want me to go look for something else, so you can ignore the fact that most of the major social and health metrics use by the UN line up better with GINI than GDP. I'm not going to do that because it's unnecessary. I already pointed out when GDP matters most: when nations are developing.

And with that we've obviously reached the end of our discussion.

Edited January 8, 2014 by cybercoma

[TimG]

We're not talking about things that DON'T correlate with GINI. I'm pointing out to you that many things do.

And you seem to be missing the entire point. I am saying that any researcher can sift though the mountains of statistics that have something to do with social well being and come up with a few that correlate. You claim that that they did not do this but my response was prove it by showing adverse data which they presented as a part of their report on correlations because if they did show the adverse data that would be a sign that they were aware of the dangers of data mining. If they did not present any adverse data then that is evidence that these correlations are likely the spurious results of a data mining effort.

You can continue to insist these correlations have some meaning but I don't see any evidence that they do at this point.

Edited January 8, 2014 by TimG

[cybercoma]

I don't need to show you adverse data to prove it. Wilkinson said clearly in his talk that they used the major UN metrics. That's your proof that they didn't just pick arbitrary statistics.

[TimG]

I don't need to show you adverse data to prove it. Wilkinson said clearly in his talk that they used the major UN metrics. That's your proof that they didn't just pick arbitrary statistics.

They used the major UN metrics that correlated. How many UN metrics did they ignore because of lack of correlation?

There are many ways to obfuscate what is fundamentally an exercise in data mining. If they did not do this then some percentage of metrics used should not have correlated. If they don't present the adverse data then it is reasonable to assume they simply selected the metrics that gave them the results they liked and ignored the ones that did not.

Edited January 8, 2014 by TimG

[cybercoma]

So they should use every last statistic gathered by the UN before they can make any claims? That's absurd.

[TimG]

So they should use every last statistic gathered by the UN before they can make any claims? That's absurd.

No it is science. They should choose in advance all stats could be said to provide 'social and health' metrics. They should then do the correlation on all such stats and report the percentage that failed to correlate at all. If this percentage is small then they may have found a real correlation. If this percentage is large then they are just data mining. If they failed to report stats that failed the correlation test then the entire study is an useless exercise in data mining.

I suspect that if I was talking about drug trials you would not be claiming that it is absurd to report results that do not support the desired outcome. In fact, you would likely demand that it be done.

Edited January 8, 2014 by TimG

[cybercoma]

They did choose them in advance. Are you even reading what I'm posting? Did you even watch the video?

[TimG]

They did choose them in advance. Are you even reading what I'm posting? Did you even watch the video?

I did not see them reporting how many stats failed the correlation test. Did I miss it? If they don't report the failure rate then I don't believe their claim that the stats were choosen in advance. So how many stats failed the correlation test? Edited January 8, 2014 by TimG

[cybercoma]

Have good night, Tim. Obviously you're not interested in having a serious discussion.

[Bob Macadoo]

I did not see them reporting how many stats failed the correlation test. Did I miss it? If they don't report the failure rate then I don't believe their claim that the stats were choosen in advance. So how many stats failed the correlation test?

Are you asking him to disprove his own claim? That must be some of the most disingenuous and/or lazy debate I've seen in awhile. I think I've finally seen all matters of trolling I could possibly imagine.

*clap*clap*clap*

Thanks Cyber your input was informative.

Edited January 9, 2014 by Bob Macadoo

[TimG]

Are you asking him to disprove his own claim? That must be some of the most disingenuous and/or lazy debate I've seen in awhile. I think I've finally seen all matters of trolling I could possibly imagine.

No - I am asking him to show that the so called correlations are not the result of data mining. One way to do that is demonstrate that the study included datasets that do not correlate. If all datasets correlate then that is a pretty good indicator that the authors are lying when they said they picked the metrics in advance and the results of the study are most likely spurious correlations. Edited January 9, 2014 by TimG

[bush_cheney2004]

Thanks... I... Huh ? How did I end up in THIS argument ?

Thank you for helping with a demonstration of the flawed income inequality meme, from the "Top 1%" to clownish "GINI coefficients" restricted only to OECD nations. The rest of the world would like to get in on the debate, but they are too distracted trying to survive on $1.00 a day.

[cybercoma]

Thank you for helping with a demonstration of the flawed income inequality meme, from the "Top 1%" to clownish "GINI coefficients" restricted only to OECD nations. The rest of the world would like to get in on the debate, but they are too distracted trying to survive on $1.00 a day.

it is not informative nor very smart to compare wealthy industrial states to countries that can't clean drinking water, don't have proper sewage, have very few hospitals/schools, etc. However, if you find it compelling to show that the US is better off than Somalia, feel free to compare.

[Michael Hardner]

Thank you for helping with a demonstration of the flawed income inequality meme, from the "Top 1%" to clownish "GINI coefficients" restricted only to OECD nations. The rest of the world would like to get in on the debate, but they are too distracted trying to survive on $1.00 a day.

Ok, since you're dragging me on to the dance floor - is wanting more a bad thing?

[bush_cheney2004]

Ok, since you're dragging me on to the dance floor - is wanting more a bad thing?

Of course not, but those hell bent on egalitarian principles and income equality sure have a very narrow perspective. Do the poor rich people in OECD nations want and need even more to get better "GINI coefficients" ?

[Moonlight Graham]

A bigger fallacy is the belief that our social problems could be solved if we only gave our wise and benevolent politicians more money to spend. Bigger government is never the solution - it is the problem.

That's probably the #1 fallacy that many conservatives/libertarians etc. make. That last statement is completely ideological rather than pragmatic. Such an absolute statement is, by its nature, easily discredited. In specific cases sometimes bigger government can be a hindrance to our society and to our economy, and in other cases increasing the size of government for a specific purpose can be beneficial. ie: Expanding government to run things like police, fire dept, the military etc. rather than private contractors is arguably a good solution. The point is, everything should be examined via a case-by-case basis, not through rigid ideological dogma.

Edited January 9, 2014 by Moonlight Graham

[Moonlight Graham]

I don't need to show you adverse data to prove it. Wilkinson said clearly in his talk that they used the major UN metrics. That's your proof that they didn't just pick arbitrary statistics.

I watched the whole video, it's interesting and I'd like to believe it, but it's still just a video with a Powerpoint presentation. It's not peer-reviewed published research. Some of the graphs don't even have values on either axis. For all your talk about research/stats methods, there's not enough hard data to chew on in the video. I need to see Wilkinson "show his work", I need to see his study, his stats, and his research methods laid out on the table, just like TimG does. Maybe Wilkinson is bang-on, maybe he's cherry-picking a lot, the point is we do know because we lack info.

[Michael Hardner]

Of course not, but those hell bent on egalitarian principles and income equality sure have a very narrow perspective. Do the poor rich people in OECD nations want and need even more to get better "GINI coefficients" ?

Indirectly, they do in that they want more for themselves. It's got nothing to do with fairness, just with 'more for me' which is as it always has been.