Reverse Discrimination Case

[jillio]

I don't know, jillio, I feel like I read somewhere that there's room for interpretation in everything....

Obviously, there is plenty of room for misinterpretation.

 

[naisho]

I still feel there are things that firefighters must know that are regardless of race prior to taking the test. If such things of normal knowledge such as basic chemistry, physics are on there, regardless of race there are things that need to be "memorized" or known as common sense. Hopefully that is what this test is all about.

One example: such as the reaction of burning aluminum (Al) + Water (H2O) + Fire, or adding a Sodium hydroxide base (NaOH) into it. Someone who doesn't know or memorize that water acts as a reactant for burning aluminum, should not become promoted for captain or chief IMO.

[ame="http://www.youtube.com/watch?v=SIUysc7VGLE"]YouTube - NaOH - Sodium hydroxide with Al in H2O makes H2 (Hydrogen)[/ame]
This is just a short video clip of explaining what I am saying. It is the reaction of NaOH (aka sodium hydroxide, a common substance in every house - found in tap water, soap, paper, construction textiles and so on) + H2O and a bit of Aluminum in it. Once the base is reacted with aluminum and water, causes gases to emit and enough to make an explosive reaction.

[ame="http://www.youtube.com/watch?v=GOEUHZRLWPs"]YouTube - Hidrogén égése- burning of hydrogen[/ame]
This is the burning example. (you have to click on the video twice to go to the actual site to see it)



Something like this is common basic chemistry knowledge that should be known regardless of racial background. If the test focuses mainly on questions related to this, then there should be no problem with determining who should get a promotion or not. A firefighter shouldn't be downing a blazing fire consisting of aluminum pieces (common in industrial areas, roofing, airplanes) with water or they're fueling it even more.

 

[Jiro]

Someone who doesn't know or memorize that water acts as a reactant for burning aluminum, should not become promoted for captain or chief IMO.

and also - the administrative responsibility, the people skill, the "connections", and the talking skill. You can be book-smart but if your people skill sucks and you are not well-respected or well-known by many, then you should not be a captain/chief either.

 

[jillio]

and also - the administrative responsibility, the people skill, the "connections", and the talking skill. You can be book-smart but if your people skill sucks and you are not well-respected or well-known by many, then you should not be a captain/chief either.

Ahhh...those things that can't be tested with pen and paper.

 

[Jiro]

Ahhh...those things that can't be tested with pen and paper.

hence oral interview..... and probably the biggest factor when deciding to pick one among the group.

To choose a minority just to meet the "requirement" or to be "fair" is illegal - in fact... a violation of federal civil law

 

[jillio]

hence oral interview..... and probably the biggest factor when deciding to pick one among the group.

To choose a minority just to meet the "requirement" or to be "fair" is illegal - in fact... a violation of federal civil law

It would appear that you are now agreeing that the test, as designed, was a poor way to test for the qualifications necessary for the job.

Maybe, maybe not, as the oral interview was a part of the test that has been determined to have problems with validity. And it was not the biggest factor according to the way the items were weighted.

This has nothing to do with choosing a minority to meet a requirement. Again, that would fall under Affirmative Action, and this case has nothing to do with Affirmative Action.

 

[Jiro]

It would appear that you are now agreeing that the test, as designed, was a poor way to test for the qualifications necessary for the job.

nope. I have not seen the test. Neither have you. Therefore - you are not right either.

Maybe, maybe not, as the oral interview was a part of the test that has been determined to have problems with validity. And it was not the biggest factor according to the way the items were weighted.

again - same comment as before.

This has nothing to do with choosing a minority to meet a requirement. Again, that would fall under Affirmative Action, and this case has nothing to do with Affirmative Action.

It's what you and daredevil have been preaching about. If the minority has not been picked for captain seat for, say.... 5 years... then you think that the test is flawed and may be racially-biased.

 

[jillio]

nope. I have not seen the test. Neither have you. Therefore - you are not right either.

Who said anything about right? I said "agree".
again - same comment as before.

Yeah, we know.:roll:


It's what you and daredevil have been preaching about. If the minority has not been picked for captain seat for, say.... 5 years... then you think that the test is flawed and may be racially-biased.

Nope, it is not what we have been preaching about. You are the one that keeps bringing up Affirmative Action, and Affirmative Action has nothing to do with what Daredevel and I are talking about. The misunderstanding continues.

 

[darkdog]

So if I understand, you would not judge a disparate impact based on percentages of whites or blacks or hispanics that pass, but based on statistical parameters like mean, SD, and skewedness. If so, how close do the means and SDs need to be between different groups? And how Gaussian does the distribution need to look? Is there some standard for that?

 

[Daredevel7]

It's what you and daredevil have been preaching about. If the minority has not been picked for captain seat for, say.... 5 years... then you think that the test is flawed and may be racially-biased.

Preaching? I'm a priest?

I am just saying, in a world where race isn't a factor, the distributions of race would be the same for the test takers and the results. We (or at least I) know that race itself should not be a factor. What makes a black man different from a white man for tests? Nothing. However, I do believe that there ARE reasons for the disparities that are related to race (low income background plus other culture effects), but I don't see how a test in real life can completely take out the culture bias... I think it's just one of those "facts of life".

I have a feeling that if you really look at the scores, the black and white people did well equally on the written test, but the black people did poorly on the oral test. That's probably why they threw out the scores. Although I think they should have gotten it assessed before they threw out the scores (the article didn't clearly state if they did).

Haven't you ever talked to someone who interacts with people differently from you due to different culture? It's obvious that white people are the majority, so imagine if you're the tester doing the oral interview, and if black people respond differently from white people, you might think "Mmm.. the white people working for him may not like how this guy works..." Not because it's the wrong answer but just a different way of doing things, and it is a very real possibility that it's true. People may not like someone running things in a different way.

 

[naisho]

I think from what Jiro mentioned earlier if the oral part is failed, that is likely leads to what he mentioned in #223 - if your oral skills suck, "better luck next time". They [the FD] generally want to choose a firefighter that has a charismatic attitude with high people skills in coaching his crewmen, good at leadership decisions, high in constitution, giving his dudes advice, what to do, man to man pep talk etc. Perhaps the whites were better at that, we don't know each and every one of them except guess at the game going on.

You also have to do a standard error for calculating deviation IIRC (if I recall correctly). But I forgot what was the formula for it.

Race might hint at a difference between different ethnicities, but the fact over general oral skills remains that it doesn't matter over the ethnicity, in the end generally organizations will choose over who could do the better job, and then that obviously discriminates the blacks putting them at a disadvantage. That's something that hints at something in this case. Wasn't one of the news articles regarding one of the suing people who was considered Dyslexic?

Now in return to their educationwise due to family poverty and the such. Is it really fair to take their family earnings and education level to match them up and play a fair game somewhere? In my burning aluminum example, it's something you either learned why it does this and why it can explode or you don't know anything and douse it all with water. So if they had horrible education, should things like this be made in exempt for them?

 

[jillio]

It's been awhile since I brushed up on socioeconomics but, isn't this a bit too low of a number to work with? The sample size is n<100 for both the test takers and the promoted.

We also do not know what is the "normal" distribution for promotions, only that this year it wasn't normal to them.

I can understand finding out the standard deviation of test scores of 77 people to find their scores and calculate what it should be, but to look at 77 persons of different races and how they ranked.. hmm?
You're also supposed to calculate a standard error for the mean to verify the validity of the standard deviation.

Another thing if I can add. We don't know who really "passed" the test. This is a crucial factor for determining validity.
We just know:
19 of 19 blacks who took the test did not get a promo.
17 of ?? whites who took the test did not get a promo.
?? of ?? hispanics who took the test did not get a promo.
?? of ?? asians / pacific islanders / native americans / other races
77 of 77 total test takers

14 whites who took the test qualified as passing.
1 hispanic who took the test qualified as passing.

What data isn't available and is probably only available to that FD/SJ, is out of the 62 other test takers, who passed and who didn't. The obviousness is that they chose people out of it, hence the comment "there is only room for 15." But you don't know if they chose 15 of the best test scorers, or if they randomly played lotto and picked whites and a hispanic, or how the dispersion of the scores were done and who was seen as promotable in their eyes.

No, this isn't too small of a sample size. This was not a sample; it was an actual population. They would only have been a sample had the test been for the purpose of norming scores or for research purposes.

Are you referring to the standard error of measurement? The mean is nothing more that the average raw score. There is no standard error of the mean. We are not looking at the normal distribution for promotions. We are looking that the distribution of scores, which in this case, was skewed and not normal. Promotion is a result of the distribution of scores, not a statisitical concept in and of itself.

One was only eligible for promotion based on specific range in test scores. Those falling below that range were not eligible for promotion. Since the purpose of the test was to determine eligibility, those falling below the range would not be considered to have "passed" the exam. Passing score would have been any score that permitted eligibility. You are attempting to put a A-D is passing, and F is failing grading system to this, and that is not what was used. If you passed you were eligible, if you didn't pass you were not.

This is not a matter of socioeconomics. It is a matter of statistics.

 

[jillio]

So if I understand, you would not judge a disparate impact based on percentages of whites or blacks or hispanics that pass, but based on statistical parameters like mean, SD, and skewedness. If so, how close do the means and SDs need to be between different groups? And how Gaussian does the distribution need to look? Is there some standard for that?

I think I know what you are trying to say, so I can only say, "Kind of." because the mean is not a statisitical parameter. It is a measure of central tendency. And in this case, the median might be a more accurrate measure of central tendency than the mean would be. Standard deviation is a measure of variability. Skewdness is type of distribution.

Your second question...I have no idea what you are asking. The distribution should approximate a normal distribution irregardless of group. If the distribution skews based on group membership, it is evidence of a bias in the instrument.

 

[jillio]

I think from what Jiro mentioned earlier if the oral part is failed, that is likely leads to what he mentioned in #223 - if your oral skills suck, "better luck next time". They [the FD] generally want to choose a firefighter that has a charismatic attitude with high people skills in coaching his crewmen, good at leadership decisions, high in constitution, giving his dudes advice, what to do, man to man pep talk etc. Perhaps the whites were better at that, we don't know each and every one of them except guess at the game going on.

You also have to do a standard error for calculating deviation IIRC (if I recall correctly). But I forgot what was the formula for it.

Race might hint at a difference between different ethnicities, but the fact over general oral skills remains that it doesn't matter over the ethnicity, in the end generally organizations will choose over who could do the better job, and then that obviously discriminates the blacks putting them at a disadvantage. That's something that hints at something in this case. Wasn't one of the news articles regarding one of the suing people who was considered Dyslexic?

Now in return to their educationwise due to family poverty and the such. Is it really fair to take their family earnings and education level to match them up and play a fair game somewhere? In my burning aluminum example, it's something you either learned why it does this and why it can explode or you don't know anything and douse it all with water. So if they had horrible education, should things like this be made in exempt for them?

By this logic, it would appear that the Society for Industrial and Organizational Psychology was right on target when they cited problems with validity. The oral portion of the test was weighted less heavily than the written portion, and that was one of the problems with the instrument that was cited. You have actually just substantiated the problems with validity.

Re: the bolded statement, that decision was already made by the courts 35 years ago.

 

[darkdog]

I think I know what you are trying to say, so I can only say, "Kind of." because the mean is not a statisitical parameter. It is a measure of central tendency. And in this case, the median might be a more accurrate measure of central tendency than the mean would be. Standard deviation is a measure of variability. Skewdness is type of distribution.

Your second question...I have no idea what you are asking. The distribution should approximate a normal distribution irregardless of group. If the distribution skews based on group membership, it is evidence of a bias in the instrument.

A few pages back, you said that "...the SD would be the same for all races in a valid test." The odds of the SDs being exactly the same are pretty close to 0, so there must be some range of acceptable difference between the SDs. I'm assuming that if the SDs have to be about the same, then the means also should be about the same. Is that a correct assumption? If so, what's the acceptable range of difference there?

Here's what I mean by the second question. If you take the mean and standard deviation of the population and throw it into the equation for a Gaussian distribution [f(x)=(population size/(SD*SQRT(2*PI)))*exp(-.5((x-mean)^2/variance))], is there a standard for how close they match up? Because a distribution can look normal to the eye without actually being normal (like a student-t). What if it has two modes? Is it more important that the skewness is close to 0? And what's the standard for how close to 0 it has to be? Because again, the odds that it's exactly 0 are pretty low.

By the way, technically speaking, mean is indeed a statistical parameter, along with standard deviation and skewness. And skewness is not a type of distribution.

 

[Daredevel7]

Found this:

New Haven Firefighters File Reverse Discrimination Lawsuit

I can tell that the person who wrote it is a bit angry about the city throwing away the scores, but it has the actual test scores. I thought it was interesting that generally hispanics do better on the written part but not the oral part and blacks do better on the oral part but not the written part......thoughts, anyone?

 

[jillio]

A few pages back, you said that "...the SD would be the same for all races in a valid test." The odds of the SDs being exactly the same are pretty close to 0, so there must be some range of acceptable difference between the SDs. I'm assuming that if the SDs have to be about the same, then the means also should be about the same. Is that a correct assumption? If so, what's the acceptable range of difference there?

Here's what I mean by the second question. If you take the mean and standard deviation of the population and throw it into the equation for a Gaussian distribution [f(x)=(population size/(SD*SQRT(2*PI)))*exp(-.5((x-mean)^2/variance))], is there a standard for how close they match up? Because a distribution can look normal to the eye without actually being normal (like a student-t). What if it has two modes? Is it more important that the skewness is close to 0? And what's the standard for how close to 0 it has to be? Because again, the odds that it's exactly 0 are pretty low.

The standard deviation of scores on a normed test is the same for anyone who takes it. That's how it is determined that one is falling outside the acceptable range of scores.

A student t is a converted score. The t score is determined by the z score. The z score is determined by the raw score. T scores and Z scores are simply formulas for raw scores that allows us to see where a score falls on a normal distribution. If all the scores have been plotted, it is obvious whether a distribution falls on a normal curve, or whether the distribution is skewed either positively or negatively. Raw scores are basically useless for determining distribution. That is why they are converted to t-scores and z-scores.

What if it does have 2 modes? That doesn't affect distribution. A normal distribution or a skewed distribution can easily have 2 modes.

A score doesn't need to fall at zero in order to skew a distribution. Few scores fall at exact zero. The far end positively or negatively of any distribution (those scores falling outside 3 SDs) will represent only .5% of any population. A majority of scores falling +/- 2 SDs from the mean will skew a distribution.

The Gaussian curve simply refers to scores plotted on a histogram that will approximate a normal distritubtion. This theory is based on the Quincunx Board developed by Galton, and is based on natural laws of the universe and explained through mathematical laws of probability. It is not a different distribution, but simply another way of demonstrating a normal distribution, and also of explaining why a skewed distibution means one needs to investigate the instrument to determine the reason for the skewness. In any normal distribution, the specific area +1SD to-1SD will include 68% of any population. +2SD to -2SD will include 95% of any population. 99.5% of the population will fall withing 3 SDs of the mean in any normal curve. These percentiles do not change on any normal distribution no matter if it is plotted using a histogram, a scatter plot, or a bell curve.

By the way, technically speaking, mean is indeed a statistical parameter, along with standard deviation and skewness. And skewness is not a type of distribution.

Normal curves and skewd curves are frequency distributions. Both the normal, the positively skewed, and the negatively skewed distribution identifies where an individual's score falls relative to the rest of the group. The mean is the most commonly used measure of central tendency and is determined by dividing the sum of X by N. Standard deviation, along with range, and interquartile range, is a measure of variability, and it is used to describe how scores vary around the mean in a standard fashion. The formulas for determining the SD is as follows:the square root of the sum of X-M squared divided by N.

Thus concludes your introductory statistics lesson for the day.

 

[jillio]

Found this:

New Haven Firefighters File Reverse Discrimination Lawsuit

I can tell that the person who wrote it is a bit angry about the city throwing away the scores, but it has the actual test scores. I thought it was interesting that generally hispanics do better on the written part but not the oral part and blacks do better on the oral part but not the written part......thoughts, anyone?

Thanks for finding the document containing both the scores and the ranks. I will run these through SPSS and come up with some additional statistics later today to confirm what you and I have been attempting to explain.

Reason for differences in oral/written scores for the black and Hispanic test takers? As we have stated all along, cultural differences in the population. Just another indication of why weighting of the items is so important to insure cross cultural validity. Supports what the Society for Insudtrial and Organizational Psychology had to say about problems in the way the sub-sections were weighted.

 

[Jiro]

Found this:

New Haven Firefighters File Reverse Discrimination Lawsuit

I can tell that the person who wrote it is a bit angry about the city throwing away the scores, but it has the actual test scores. I thought it was interesting that generally hispanics do better on the written part but not the oral part and blacks do better on the oral part but not the written part......thoughts, anyone?

again - this doesn't warrant any microscopic analysis or any change. simple - it depends on how the Fire Commission chiefs want it. If they feel a need for book-smart chief, go ahead and pick the one who did well on written part. If they feel the team could use some leadership and pep talk, then pick someone who talks good.

and what's there to bitch about? life's never fair. there are plenty of CEO's and management positions that have been filled by under-qualified people. Ever heard of "Office Politic" ?

 

[jillio]

again - this doesn't warrant any microscopic analysis or any change. simple - it depends on how the Fire Commission chiefs want it. If they feel a need for book-smart chief, go ahead and pick the one who did well on written part. If they feel the team could use some leadership and pep talk, then pick someone who talks good.

This makes absolutely no sense in light of the fact that eligibility was determined based on test scores.