But New Hampshire would be the first to pass such a bill, and for good reason, said Matt Mincieli, the Northeast executive director of TechNet, a technology advocacy organization that includes companies like Apple, Amazon, Facebook, Microsoft, Verizon and Uber. Bills in other states, including Massachusetts, have gotten no traction.
“This bill would be a gift to cybercriminals,” Mincieli wrote in the Concord Monitor, and circulated the article to the lawmakers. And he testified at the hearing, “No one really knows once we open up tech to anyone. Who would protect your security?”
The bill would scare technology companies away from New Hampshire, he insisted, and it’s so broad that it would affect nearly every type of industry.
The bill, he noted, didn’t just ask companies to reveal schematics, but also diagnostic software, services access passwords, updates and corrections to firmware (though advocates said it wouldn’t reveal encrypted security information).
Indeed, the bill drew testimony from beyond usual technology suspects: securities companies, the entertainment and gaming industry and even farm equipment.
William Taranovich, Jr., president of North Country Tractor, an authorized John Deere dealer from Pembroke, testified brandishing a long screwdriver and a laptop. In 1991, he mainly used the former to fix equipment, but today even chainsaw repairs need proprietary software, “so the saw won’t jump up and hit you in the face.”
Apple, Amazon, Facebook, John Deere, Microsoft, Verizon and Uber I am somewhat terrified that John Deere recommends maintenance on operating machinery. source: TechNet report
Is the max ipad plus. Obviously that needs to come next year, but we can dream, because the ipad plus max just sounds terrible.
Also, just to ask: how do we pluralise the iphone Maxs? As the MaxSS? I suppose that we can be thankful that the Lightning connector is being removed, since they would have to do something clever, like denote the SS with lightning bolts...
Invisible threats are more unnerving than ones we're familiar with. Pools killing thousands per year? Meh. It's a pool. People falling off roofs? Well, it happens.
I think the correct response is to have nuclear power kill more people. Then everything will be normalized.
As one of those grad students once upon a time: 95% of the social scientists have absolutely no clue about the tools they use. They do things because they are traditional checkboxes, and those are the marks reviewers look for. They don't care about the logic of the tools they use, and as a result, shit gets dumped out without comprehension by either reviewers or authors.
The more prime numbers that are discovered, the more likely we are to be able to discover a pattern within an arbitrary base number set. The larger numbers are useful because we also want to make sure that the entire range is consistent, or in other words that any pattern, or lack of pattern, is the same across the entire set of numbers. There is always a benefit to trying to find patterns in number theory -- it's one of the coolest and most interesting fields in pure mathematics.
1. Mathematical foundations of infinite-dimensional statistical models 2. Asymptotics in statistics: Some basic concepts 3. Real Analysis 4. Real Analysis and probability 5. Asymptotic methods in statistical decision theory
One of the interesting things about all (or almost all) ``Clinical Scales'' is that they are all equally weighted when the scale/test is being scored. The difficulty and discrimination differences of each item are completely scrapped in favour of the a constant scaling of sum_{i=1}^{p}1*response_{i}. This is unfortunate, and seeing the ``clinically validated'' bit got me thinking about a fun study I could do. A very quick review however found this: An item response theory evaluation of three depression assessment instruments in a clinical sample Mats Adler, Jerker Hetta, Göran Isacsson and Ulf Brodin[0] an open access and from a quick glance, fairly well done, demonstrating that the conventional scaling is wrong. I have some personal differences I would have applied, but it happens.
Hell yes. I run the beta versions, and bug report when I can and only use FLOSS software like an addict. I admit my ethical preference, and dealing with a Psychology department as often as I do, it's an immense problem that often gets me labelled a zealot. I prefer LaTeX and use point out mathematical flaws and software default errors that result in false logical errors in colleagues' papers like an immense jackass.
However, even I have to admit that using Chromium, with all its Googley evilness (yes Chromium != Chrome) is much, much faster and responsive than Firefox is. That should worry the mother fucker out of the Mozilla developers, that a hard core FLOSS zealot is willing to admit superiority in another suite of software. Does it however? Apparently not as much as abandoning Thunderbird and focusing on the latest visual makeover for the Mozilla brand.
Cleaner likely refers to the fact that they only conducted a standard OLS regression, rather than a hierarchical model in which the commonalities between the multiple sites are properly modelled.
It's lazy, but fairly common in the social sciences: if the repeated identical dimensions aren't properly modelled (i.e., are left as unique repetitious observations) the standard errors for each of the effects being estimated are shrunk, leading to Type 1 (false positive) errors.
It should be a trivial logical argument to suggest that "more accurate information leads to less error generated behaviour." One should note that the clause "accurate information" is a hell of stated requirement, but the measurement estimation of accurate true scores from error contaminated dimensions is a well studied field. Dealing with that in a online manner is more difficult, but a growing field.
Very courteous reply, and always happy to talk statistics. I miss the old Slashdot.
Anyway, assuming a normal distribution, 50% of the distribution is below the mean (integral from -Inf to 0, assuming we've centred the distribution at 0). Basically what they're deriving from those numbers is a normal distribution with mean of 100 and sd of 15 (and therefore variance=225). The first two moments norm and variance define the distribution of a Gaussian variable, and from this we know the portion of the probability curve under the density function with the help of the integral, which is the source of the mean plus/minus one sd is a 34 point confidence interval around the mean on either side (i.e., 68% overall). 95% being 100+/-1.96*(15),
The problem, I think, was you were deducing from the centre of a symmetric distribution, and so the proportion you're trying to fit around is the middle of the chunk, not the bottom proportion. I may have always misinterpreted your intention, and if so I apologise, but on rereading, I think my original interpretation was still correct. Anyway, the true answer would be that in a normal distribution, the mean is equal to the median, and so 50% is below the average value, by definition.
15% of the population is below average intelligence? Speaking as a psychometrician, our models assume a latent intelligence space that is normally distributed (actually, while most of the item response models are parametric, I personally prefer mokken scaling, or ordinal IRT. Both however assert stochastic ordering for dichotomous items) which would make the realisation of 15% of the area under the curve impossible for the integral of scores from (-Inf,0). If 15% is the actual area below the mean score IQ, then the IQ score isn't normally distributed, and both IRT and classical true score theories are wrong (and we knew that true scoroe theories, what most of psychology is based upon, were wrong) as solved for now. In either case, either the value you claim is wrong, or the derivation of the model upon which your claims are built are wrong. Clay feet nd all.
Nein. Consider how many scales are "validated" by Psychology post-docs, either with PCA that they call factor analysis or confirmatory factor analysis. In both cases the measurement structures are wrong (non-continuous data --> a Hessian that's overly inflated, and therefore presumed to be more informative that reality holds): PCA only works for formative measurement structures (i.e., the items cause the latent structure, like the Hollingshead SES scale), whereas nearly all social science measurement assumes a reflexive structure (underlying cause which is reflected by the common answering patterns across items), and can be easily attested to if one reports the Cronbach's alpha measure (a ratio of the signal variance to the total variance: although note that variance is undefined for ordinal variables, from a measurement, not predictive, viewpoint) in the paper. Notice the right wing authoritarianism scale, which claims to be tri-dimensional, with strong Cronbach's alpha reliability, also uses the dimensionality drawn from PCA, on ordinal items, which are mutually exclusive claims, but have never been critiqued. Nearly all scales are drawn up in this fashion. The mathematical test of confirmatory factor analysis actually imposes the reflexive structure, and allows for one to associate specific items with the assumed latent causes: however, the test assumes a multivariate normal joint distribution across the observed information, which is almost never seen in Psychology. This is the reason why the likelihood-ratio test, whose null hypothesis is that the model as specified by the researcher is a good approximation to the data, is nearly universally ignored despite being almost universally found to be significant in applications (look at the kvetching of Les Hayduk about this issue). Part of the issue is the assumption that the items are continuous, which again inflates the information in each item, and misrepresents the nature of the data. Item response theory (or more specifically, multidimensional item response theory) perfectly solves these issues, but it is considered a niche tool for us psychometricians, who can be safely ignored for giving a fuck about what the data says. See the responses of Susan Fiske, an endowed chair Princeton PhD of psychology, as related by Gelman [0].
When confronted with errors in their mathematical/statistical analysis of their results (making invalid assumptions that were empirically unjustified to validate the use of traditional OLS regression), I had two different professors give me a response:
I am not a statistician, I'm a psychologist.
That is a paraphrase of one, and a verbatim of the other. My final year as a undergraduate, my school offered a `psychometrics and testing' course, which deals with the mathematical theory of relating imperfect observations to actually measurement inferences. This professor, a clinician, intended the end of the first class to be a bragging ritual of what a great methodological expert he was. He got 10 minutes into it before I raised my hand and started correcting the gross errors that he was espousing to the students. Basic statistical operation facts, one of the more prominent ones being that the standard assessment of adequacy for a simultaneous (structural) equation model in reproducing observed data is to produce a chi**2 value (df=free parameters) with a p-value less than.05
This is actually the opposite of how the method works, but he defended that he was an expert and knew that of which he spoke.
Now, I'm a second year graduate student specialising in psychometrics and quantitative psychology. I finished two masters degrees in one year, and am taking my comprehensives in the Spring, because I couldn't do it any faster with the administrative bureaucracy. My dissertation work is in relaxing the multivariate normality assumptions that plague simultaneous equations, which are solved by using multivariate normal maximum likelihood. I work with information in statistical practice as defined as the inverse log(probability) (or entropy), viewed as the amount of error in our data that we cannot predict.
I am an extreme example of my subfield of psychology (normally, when we attempt to publish methodological developments and maths heavy papers in the typical Psychology study journals, we get told to publish them in the statistical methods journals, because those are who will understand them. These papers are rejected at the editorial level, and not even sent out to be peer reviewed. The fact that the actual level of competence in their readership is so low doesn't seem to be worth contemplating). I actually consult on national education projects, private industry scale development (autism scale development, for example), and I bring in 200% of what my school pays me as a stipend in personal funding, plus the actual study funding. As an undergraduate, I brought in such a large amount of funding, that you had to go back almost thirty years for the ENTIRE DEPARTMENT to produce the same amount of money. Their defence was that they were a teaching institution and research was secondary.
The problem is psychologists don't seem to understand the maths behind the p-values they report (and confidence intervals, while heralded as the salvation of relying on p-values, are the exact same thing). They don't understand that the assumptions that underlie the methods they use are not universal constants that can just be taken as the divining rods of truth. I've actually criticised people's dissertations that were on the cusp of defence, as not even being remotely statistically valid and yet they've gone on to a successful defence without changing a damn thing. It is just a variation of sticking your head into the sand.
Of course this isn't simply relegated to psychology, or even just the social sciences as others have pointed out. It is endemic to them though. Three positions were created as `quantitative methods' experts at my undergraduate school that were filled in my final year. One was filled by a statistician with the maths department, and the two of us are working on creating a statistics focused `Quantitative Social Sciences' program. One was a psychologist, who I, of course, questioned when they were brought in for and who answered the bulk incorr
You should really sign in when posting something so informative (although some citations would be really nice). That way you can get good karma to give your posts an automatic +1 or +2 and it's easier to track when people reply.
fair point. I lost my old account a good six years ago and never created a new one.
Slashdot Mirror
But New Hampshire would be the first to pass such a bill, and for good reason, said Matt Mincieli, the Northeast executive director of TechNet, a technology advocacy organization that includes companies like Apple, Amazon, Facebook, Microsoft, Verizon and Uber. Bills in other states, including Massachusetts, have gotten no traction.
“This bill would be a gift to cybercriminals,” Mincieli wrote in the Concord Monitor, and circulated the article to the lawmakers. And he testified at the hearing, “No one really knows once we open up tech to anyone. Who would protect your security?”
The bill would scare technology companies away from New Hampshire, he insisted, and it’s so broad that it would affect nearly every type of industry.
The bill, he noted, didn’t just ask companies to reveal schematics, but also diagnostic software, services access passwords, updates and corrections to firmware (though advocates said it wouldn’t reveal encrypted security information).
Indeed, the bill drew testimony from beyond usual technology suspects: securities companies, the entertainment and gaming industry and even farm equipment.
William Taranovich, Jr., president of North Country Tractor, an authorized John Deere dealer from Pembroke, testified brandishing a long screwdriver and a laptop. In 1991, he mainly used the former to fix equipment, but today even chainsaw repairs need proprietary software, “so the saw won’t jump up and hit you in the face.”
Apple, Amazon, Facebook, John Deere, Microsoft, Verizon and Uber
I am somewhat terrified that John Deere recommends maintenance on operating machinery.
source: TechNet report
Is the max ipad plus. Obviously that needs to come next year, but we can dream, because the ipad plus max just sounds terrible.
Also, just to ask: how do we pluralise the iphone Maxs? As the MaxSS? I suppose that we can be thankful that the Lightning connector is being removed, since they would have to do something clever, like denote the SS with lightning bolts...
Invisible threats are more unnerving than ones we're familiar with. Pools killing thousands per year? Meh. It's a pool. People falling off roofs? Well, it happens.
I think the correct response is to have nuclear power kill more people. Then everything will be normalized.
If only higher math was useful. Just seems like a giant circle jerk.
I, very strongly, recommend consulting Abstruse Goose. Be sure to check the hover text:
There is no branch of mathematics, however abstract, which may not some day be applied to phenomena of the real world.
Space is big, dude. Do you know how many paperclips you can make out there?
42?
Shouldn't they have seen it coming if they were any good at their jobs?
As one of those grad students once upon a time: 95% of the social scientists have absolutely no clue about the tools they use. They do things because they are traditional checkboxes, and those are the marks reviewers look for. They don't care about the logic of the tools they use, and as a result, shit gets dumped out without comprehension by either reviewers or authors.
The more prime numbers that are discovered, the more likely we are to be able to discover a pattern within an arbitrary base number set. The larger numbers are useful because we also want to make sure that the entire range is consistent, or in other words that any pattern, or lack of pattern, is the same across the entire set of numbers. There is always a benefit to trying to find patterns in number theory -- it's one of the coolest and most interesting fields in pure mathematics.
1. Mathematical foundations of infinite-dimensional statistical models
2. Asymptotics in statistics: Some basic concepts
3. Real Analysis
4. Real Analysis and probability
5. Asymptotic methods in statistical decision theory
One of the interesting things about all (or almost all) ``Clinical Scales'' is that they are all equally weighted when the scale/test is being scored. The difficulty and discrimination differences of each item are completely scrapped in favour of the a constant scaling of sum_{i=1}^{p}1*response_{i}. This is unfortunate, and seeing the ``clinically validated'' bit got me thinking about a fun study I could do. A very quick review however found this: An item response theory evaluation of three depression assessment instruments in a clinical sample Mats Adler, Jerker Hetta, Göran Isacsson and Ulf Brodin[0] an open access and from a quick glance, fairly well done, demonstrating that the conventional scaling is wrong. I have some personal differences I would have applied, but it happens.
[0] https://bmcmedresmethodol.biom...
INGREDIENTS PER 473ml/ 16 Fl. Oz.: Caffeine 131mg. You may want to reread that.
Hell yes. I run the beta versions, and bug report when I can and only use FLOSS software like an addict. I admit my ethical preference, and dealing with a Psychology department as often as I do, it's an immense problem that often gets me labelled a zealot. I prefer LaTeX and use point out mathematical flaws and software default errors that result in false logical errors in colleagues' papers like an immense jackass.
However, even I have to admit that using Chromium, with all its Googley evilness (yes Chromium != Chrome) is much, much faster and responsive than Firefox is. That should worry the mother fucker out of the Mozilla developers, that a hard core FLOSS zealot is willing to admit superiority in another suite of software. Does it however? Apparently not as much as abandoning Thunderbird and focusing on the latest visual makeover for the Mozilla brand.
Sigh.
Cleaner likely refers to the fact that they only conducted a standard OLS regression, rather than a hierarchical model in which the commonalities between the multiple sites are properly modelled.
It's lazy, but fairly common in the social sciences: if the repeated identical dimensions aren't properly modelled (i.e., are left as unique repetitious observations) the standard errors for each of the effects being estimated are shrunk, leading to Type 1 (false positive) errors.
It should be a trivial logical argument to suggest that "more accurate information leads to less error generated behaviour." One should note that the clause "accurate information" is a hell of stated requirement, but the measurement estimation of accurate true scores from error contaminated dimensions is a well studied field. Dealing with that in a online manner is more difficult, but a growing field.
I saw a documentary about the Third World of the Antares system on Futurama.
Very courteous reply, and always happy to talk statistics. I miss the old Slashdot.
Anyway, assuming a normal distribution, 50% of the distribution is below the mean (integral from -Inf to 0, assuming we've centred the distribution at 0). Basically what they're deriving from those numbers is a normal distribution with mean of 100 and sd of 15 (and therefore variance=225). The first two moments norm and variance define the distribution of a Gaussian variable, and from this we know the portion of the probability curve under the density function with the help of the integral, which is the source of the mean plus/minus one sd is a 34 point confidence interval around the mean on either side (i.e., 68% overall). 95% being 100+/-1.96*(15),
The problem, I think, was you were deducing from the centre of a symmetric distribution, and so the proportion you're trying to fit around is the middle of the chunk, not the bottom proportion. I may have always misinterpreted your intention, and if so I apologise, but on rereading, I think my original interpretation was still correct. Anyway, the true answer would be that in a normal distribution, the mean is equal to the median, and so 50% is below the average value, by definition.
15% of the population is below average intelligence? Speaking as a psychometrician, our models assume a latent intelligence space that is normally distributed (actually, while most of the item response models are parametric, I personally prefer mokken scaling, or ordinal IRT. Both however assert stochastic ordering for dichotomous items) which would make the realisation of 15% of the area under the curve impossible for the integral of scores from (-Inf,0). If 15% is the actual area below the mean score IQ, then the IQ score isn't normally distributed, and both IRT and classical true score theories are wrong (and we knew that true scoroe theories, what most of psychology is based upon, were wrong) as solved for now. In either case, either the value you claim is wrong, or the derivation of the model upon which your claims are built are wrong. Clay feet nd all.
Nein. Consider how many scales are "validated" by Psychology post-docs, either with PCA that they call factor analysis or confirmatory factor analysis. In both cases the measurement structures are wrong (non-continuous data --> a Hessian that's overly inflated, and therefore presumed to be more informative that reality holds): PCA only works for formative measurement structures (i.e., the items cause the latent structure, like the Hollingshead SES scale), whereas nearly all social science measurement assumes a reflexive structure (underlying cause which is reflected by the common answering patterns across items), and can be easily attested to if one reports the Cronbach's alpha measure (a ratio of the signal variance to the total variance: although note that variance is undefined for ordinal variables, from a measurement, not predictive, viewpoint) in the paper. Notice the right wing authoritarianism scale, which claims to be tri-dimensional, with strong Cronbach's alpha reliability, also uses the dimensionality drawn from PCA, on ordinal items, which are mutually exclusive claims, but have never been critiqued. Nearly all scales are drawn up in this fashion. The mathematical test of confirmatory factor analysis actually imposes the reflexive structure, and allows for one to associate specific items with the assumed latent causes: however, the test assumes a multivariate normal joint distribution across the observed information, which is almost never seen in Psychology. This is the reason why the likelihood-ratio test, whose null hypothesis is that the model as specified by the researcher is a good approximation to the data, is nearly universally ignored despite being almost universally found to be significant in applications (look at the kvetching of Les Hayduk about this issue). Part of the issue is the assumption that the items are continuous, which again inflates the information in each item, and misrepresents the nature of the data. Item response theory (or more specifically, multidimensional item response theory) perfectly solves these issues, but it is considered a niche tool for us psychometricians, who can be safely ignored for giving a fuck about what the data says. See the responses of Susan Fiske, an endowed chair Princeton PhD of psychology, as related by Gelman [0].
[0]. http://andrewgelman.com/2016/0...
When confronted with errors in their mathematical/statistical analysis of their results (making invalid assumptions that were empirically unjustified to validate the use of traditional OLS regression), I had two different professors give me a response:
I am not a statistician, I'm a psychologist.
That is a paraphrase of one, and a verbatim of the other. My final year as a undergraduate, my school offered a `psychometrics and testing' course, which deals with the mathematical theory of relating imperfect observations to actually measurement inferences. This professor, a clinician, intended the end of the first class to be a bragging ritual of what a great methodological expert he was. He got 10 minutes into it before I raised my hand and started correcting the gross errors that he was espousing to the students. Basic statistical operation facts, one of the more prominent ones being that the standard assessment of adequacy for a simultaneous (structural) equation model in reproducing observed data is to produce a chi**2 value (df=free parameters) with a p-value less than .05
This is actually the opposite of how the method works, but he defended that he was an expert and knew that of which he spoke.
Now, I'm a second year graduate student specialising in psychometrics and quantitative psychology. I finished two masters degrees in one year, and am taking my comprehensives in the Spring, because I couldn't do it any faster with the administrative bureaucracy. My dissertation work is in relaxing the multivariate normality assumptions that plague simultaneous equations, which are solved by using multivariate normal maximum likelihood. I work with information in statistical practice as defined as the inverse log(probability) (or entropy), viewed as the amount of error in our data that we cannot predict.
I am an extreme example of my subfield of psychology (normally, when we attempt to publish methodological developments and maths heavy papers in the typical Psychology study journals, we get told to publish them in the statistical methods journals, because those are who will understand them. These papers are rejected at the editorial level, and not even sent out to be peer reviewed. The fact that the actual level of competence in their readership is so low doesn't seem to be worth contemplating). I actually consult on national education projects, private industry scale development (autism scale development, for example), and I bring in 200% of what my school pays me as a stipend in personal funding, plus the actual study funding. As an undergraduate, I brought in such a large amount of funding, that you had to go back almost thirty years for the ENTIRE DEPARTMENT to produce the same amount of money. Their defence was that they were a teaching institution and research was secondary.
The problem is psychologists don't seem to understand the maths behind the p-values they report (and confidence intervals, while heralded as the salvation of relying on p-values, are the exact same thing). They don't understand that the assumptions that underlie the methods they use are not universal constants that can just be taken as the divining rods of truth. I've actually criticised people's dissertations that were on the cusp of defence, as not even being remotely statistically valid and yet they've gone on to a successful defence without changing a damn thing. It is just a variation of sticking your head into the sand.
Of course this isn't simply relegated to psychology, or even just the social sciences as others have pointed out. It is endemic to them though. Three positions were created as `quantitative methods' experts at my undergraduate school that were filled in my final year. One was filled by a statistician with the maths department, and the two of us are working on creating a statistics focused `Quantitative Social Sciences' program. One was a psychologist, who I, of course, questioned when they were brought in for and who answered the bulk incorr
and yes, dammit, I know I should have logged in before posting.
You should really sign in when posting something so informative (although some citations would be really nice). That way you can get good karma to give your posts an automatic +1 or +2 and it's easier to track when people reply.
fair point. I lost my old account a good six years ago and never created a new one.