Why Placing Suicide Barriers on the Cold Spring Bridge Is Not the Answer

I reach a different conclusion about the Seiden report, after reading it and discussing it with experts.

The conclusion is crystal clear from Seiden: some lives are saved by a barrier. It is not reasonable to claim that *all* of the 515 suicidal folks who were found on the Golden Gate Bridge and subsequently analyzed by Seiden were not pertinent to the consideration of barriers, which is what both you and Garrett Glasgow maintain.

Your claim about `no decrease in the overall suicide rate' isn't helpful, because neither you nor Garrett Glasgow or anyone else has been able to device a statistical technique that *could ever* have sensitivity to the decrease due to prevention of bridge suicides. Confounding variables and lack of sufficient statistics preclude such sensitivity. I don't know why you and Garrett Glasgow don't do the pertinent arithmetic to evaluate the sensitivity. Since you don't bother to evaluate your sensitivity, but then go on to argue as if the results that have been obtained are still relevant you look either ill-informed or insincere.

As for the money, you have somehow inflated $605,000 to $1,000,000. Why? In any case, you make no comparison of this number to the ongoing expense on Mental Health services in this County, which is at least $1 billion over the lifetime of the proposed barrier.

For every $1000 spent on counseling and other Mental Health services, then, less than $1 would be spent on the bridge barrier. Why is that unreasonable?

As for the beauty and aesthetics, that is purely subjective. Some people might be far more in awe of the bridge if they knew it was no longer a suicide magnet.

What do these costs of Mental Health services have to do with anything here?

Would anyone be surprised if a Caltrans project initially estimated at costing $605 thousand at the beginning ends up costing an actual $1 million at the end?

For $1 million, what else could Caltrans do to improve safety and save lives in as small an areas as Santa Ynez Valley? How about road improvements or mandatory breath alcohol testers installed at Rancho San Marcos Golf Party House or the Casino?

1)The public funding for Santa Barbara County's Mental Health Services is a measure of what is now spent on the types of services that Dr. McGinnes shifting the $605,000 to. That spending really dwarfs the cost of the barrier, by over a factor of 1,000. It seems difficult to accept that changing the Mental Health Services by about 0.1% would save as many lives as the Cold Spring bridge barrier would save; certainly no-one has offered an analysis of whether the $1 billion spent every 30 years on County Mental Health services is being spent effectively, or whether 1% of the funding there (10 times the cost of the bridge barrier) could be redeployed more effectively to prevent suicide. Additionally, Dr. McGinnes does not provide a practical mechanism whereby $605,000 of Caltrans money could be transfered to the activities he would prefer to see money spent.

2)The estimate I'm aware of is $605,000; it would be nice if Dr. McGinnes could explain how he got up to $1 million. To inflate the number without explanation might suggest that Dr. McGinnes is a bit careless with details... does a similar absence of care pervade his other analyses?

3)Caltrans is always conducting safety analyses, and lots of road improvements have been made on 154 over the past 10 years; there are more road separations, for example. The most effective activity, I think, would be heavy ticketing of speeders on 154.

4)Nothing is preventing the Casino or the Golf Party house from installing those items tomorrow.

Right, so if they put a barrier on one side of the bridge do you think it will reduce suicide by 50% ?

I don't think some of you really thought through what you are saying.

Your idea of a barrier on 1/2 of the bridge has never been brought up, and it is not likely to reduce suicides from the bridge, because people will cross the bridge and jump off the other side.

The Seiden study of 515 suicidal folks found on the Golden Gate Bridge indicated that once off the bridge, they generally did *not* go on to commit suicide by other means. Your idea does not get them off the bridge before the temptation is removed.

As I said in my essay, I and others have carefully studied the CalTrans proposal and the studies it cites.

Clearly set forth in black and white in the document which purports to grant a conceptual approval of the proposal is the figure of $1,000,000.

This document is a public record, so you can take a look at it for yourself, sevendophins. Or I can show you the copy I obtained as a part of my careful research.

Dr. Mcginnes,

The cost I've seen linked to the Caltrans web page on the topic -

http://www.dot.ca.gov/dist05/projects/sb...

is on page 5 of:

http://www.dot.ca.gov/dist05/projects/sb...

where it says at the bottom:

Project Construction Cost: Estimated $605,000

can you provide a link to the report where $1,000,000 is quoted?

To "sevendolphins":

As I said in closing in my essay above, "In the spirit of the principle that 'understanding does not require agreement,' I sincerely look forward to participating in the respectful and well-informed dialogue that is needed to work out a wise decision regarding the Caltrans proposal in the weeks and months to come."

Responding to your last comment, then:

With respect, "Sevendolphins",you cannot be sufficiently well-informed about this or any other CalTrans proposal if you do not undertake careful research, and that means going further than merely looking at the CalTrans website. Pursuant to the California Public Records Act, I obtained a copy of an internal CalTrans memorandum dated 12 September 2006 from Janice Benton to
Paul McClintic ("re: File 3.1.20 / 05-SB-154") wherein the $1,000,000 is clearly set forth.

Thank you for the opportunity to be of assistance to you about this matters or dollars and cents.

In the wider sense, I hope you will try to understand why I and many, many others so strongly opposed to this proposal. To that end, I close with a few lines from another essay I was asked to write about this matter:

Poor project planning such as this must be confronted head-on and prevented from wasting public monies on things that do not effectively address legitimate public needs. Remember that it was CalTrans who also tried to force our community to accept an elevated freeway all along the downtown stretch of 101 and how we had to struggle mightily to prevent that ill-considered scheme.

Last but not least, it is important for us to take into consideration the breathtakingly spectacular beauty of the Cold Spring Canyon Arch Bridge and to be mindful of the beautiful experience enjoyed by the thousands upon thousands of people who look at and pass over it. It a remarkable gem along our justly famed Scenic Highway, and not just “a stupid old bridge” as a speaker at recent Glendon forum described it, and we should not carelessly deprive ourselves of its inspiration.

A friend put it this way, and I add my voice to hers:

"The suicide of anyone makes everyone feel bad, and yet this must not lead us to deprive ourselves of the grace and beauty of places and experiences that sustain our spirits as we carry on from there.”

Thank you, Dr. McGinnes, perhaps you can post that document online so the entire context of the discussion and meaning of the $1,000,000 can be deeply understood by all of us.

For example, it may be Janice Benton's personal opinion is that the cost is $1,000,000, and that the consensus of the entire estimation process at Caltrans is $605,000.

As much as you say you have done careful research, you have not stated a reason why the publicly avaliable documents state $605,000, while one internal memo states $1,000,000. Reconciling those numbers is the essence of deep understanding, and since I'm sure your research has been so careful as to uncover what the precise differences are in the accounting that leads to the two numbers, it would be ever so helpful of you to write down the reconcilliation here.

From my perspective, the comparison of a barrier along the Cold Springs Bridge and the 101 through Santa Barbara is not even comparing apples and oranges, nor even comparing a blueberry and a watermelon. Whether or not 1/2 mile of freeway is raised 15-20 feet is just a far bigger deal than a nearly-transparent barrier along a bridge where pedestrian traffic is forbidden, and most people are driving across at 50 mph anyway.

It would of course be very helpful and demonstrate your care if you could post your estimates of statistical sensitivity inherent in the Glasgow study.

The negative correlation between suicide rate and number of bridges in the Glasgow study is enormous, and far larger than the entire number of bridge suicides. That negative correlation cannot have anything to do with bridge suicides.

The dispersion of the data around the trendline is so large as to preclude any sensitivity to bridge suicides. In the end, that means that the Glasgow study never had and never will have any statistical ability to add to the world's knowledge on the efficacy of barriers. And, I think, a careful and thoughtful study would have clearly stated that fact.

Here's another option:

Ban bridges. Did you know 100% of people who die after jumping of a bridge jumped off a bridge?

Clearly, "removal of temptation" is an asinine reason to spend $600k. Should we remove all temptation? No more Nirvana on the radio, ban Romeo and Juliet?

Maybe the money could be spent, hiring people to watch the bridge 24/7. They could be trained in psychology and suicide-prevention techniques, and be allowed arrest-powers, for extreme cases. They could be called "catchers", and be deputized by the Department of the Rye. Arrestees would be put to community service, comforting patients of fatal-diseases, like in the AIDS ward of a local hospital, or a ward of terminally ill children.

Or, there could be a base-jumping booth set up by the bridge, for those who not be 100% decided on jumping, and might need a test run. To weed out the thrill-seekers, potential jumpers would be required to sit in on an anti-suicide meeting, and participate in "trust" exercises, where they catch others, falling from a fake bridge. It might do everyone involved some good.

As a third option, a high-stakes game of Russian Roulette could be arranged for PJ's (Potential Jumpers). If they don't die, they might find some meaning to life, or at least something to do with all the money that they win (for awhile).

And before I forget anyone who might actually have a mental illness, how about a coin-operated, video-recording camera machine, so at least there would be a record, of who these people are (were), and maybe why they felt the need to jump from the bridge?

Any of these should cost well under $600K, although they may incur ongoing costs.

A cheaper, and possibly more effective idea (not necessarily a "solution"), might simple to post volunteers at the bridge, so that there would be someone to listen, when a PJ shows up. Sometimes, that's all that's needed. . . .

The barrier will last something like 30 years. That works out to $20,000 a year; I doubt you could post someone there for a useful fraction of time on that amount of yearly money. BTW, the Mental Health portion of the County budget is $41 million a year; perhaps they could post someone.

I guess some people enjoy making humor out of suicide. To each their own. Some people find it humorous to torture animals. Others general find humor in the pain of other people, and so there is a whole genre of pornography devoted to snuff, forced sex, etc. It's a free country, with a free press, but I don't find the decisions made by such people particularly appealing.

If you could actually weigh the effectiveness of a proposed solution, against the cost of it, would that make a difference? A cheap solution that really isn't a solution, is still more expensive than one that works. Of course, trying to reconcile spending money to save lives in this situation is really kind of ludicrous--there are no guarantees.

BTW, with the exception of those whom we could classify as truly, mentally ill, I think that most people contemplating ending their lives really do simply need someone to listen, and maybe give a sh!t--pardon the phrase. I've know a few people who have needed and ear, and I've also seen the after-effects of what happens when someone leaves the questions of their death unanswered. [One caveat, it may take more than listening to someone once, as thoughts of suicide may be cyclical, approaching chronic.]

Finally, using humor--insensitive as it may seem--is sometimes the only way to get a different perspective on an issue.

In general, however, if someone is has dedicated themselves to ending their own life, they won't be stopped. And, referencing the post above, about those not jumping from the Golden Gate bridge, perhaps they simply had not despaired enough. --I'm not saying that they should have, but that they simply didn't pass the point of no return.

Oh, and lastly, I wholly believe that one should retain the right to end one's own life, I don't beleive that most of us have a good reason to--Kevorkian consultations notwithstanding.

I think Drs. McGinnes, Dr. Glasgow, and the last poster seem to believe that *none* of the 515 folks talked down off the Golden Gate Bridge and subsequently tracked by Dr. Seiden were serious about suicide. That is really hard to believe.

I have no doubt that a barrier on the Cold Spring Bridge would stop people from jumping off the Cold Spring Bridge. Combined with the Seiden study, I have no doubt that some of those people will not go on to find other means. Seems to me that qualifies as effective.

BTW, the lost wages from the suicide, family members, social services, etc. amount to about $1.5 million per suicide, according the the national center for the prevention of suicide. Taxes being about 30%, about two suicides stopped pays for the barrier at least from the narrow perspective of tax monies.

I think suicide is in most cases (not all) a disease, a permanent solution to a temporary problem. There are people who work out all the details including their own body disposal, or who are old and infirm or in terrible pain, but they are a tiny minority. The great majority of suicides are preventable, IMO, and the would-be suicides are quite happy that they didn't do it. The dark, sadistic, adolescent humor of some people doesn't help at all in preventing them, just as humor about rape doesn't do anything to prevent rape, or humor about lynching doesn't do anything to prevent hate crime. There is freedom of speech in this country but that doesn't mean that all speech makes ours a better country.

Sevendolphins, methinks you doth assume too much.

Not that it matters, but I generally agree with your comment, "a permanent solution to a temporary problem". And, if you care to see, that feeds into what I posted about "the point of no return". I wasn't saying that those who didn't kill themselves weren't serious--that's your slant--I'm saying that they hadn't passed the point where they couldn't regret it. . .at least, not in *this* life. In any case, you won't normally catch me using a blanket, "ALL", for a group of people, in any situation, as that's just not possible--or, at least, likely.

I hope that you might be able to understand that, counterpoint to those that are glad they didn't jump from the Golden Gate, there might be some that still wish they had. Unfortunately, we don't have that data, nor are we able to interview those who completed their journey.

BTW, looking at the barrier logically, without regard to the monetary issue, you could say that if bridge barriers save lives, then they are a good thing. However, I still think that it's a better thing by far, for us as a human family, to be attentive to those around us, and let that be the support mechanism, for curtailing suicides.

And, if you pardon the tangent, if we take on the burden of saving people from their desires to end their lives quickly, regardless of the consequence(s), then are we to become our Brother's keeper in all things?

And. . .yet. . .finally: I read into your posts, that you've gained a distaste for what you have judged to be my character, but I hope that you're able to let that go. Comedy and tradgedy are often bedfellows, my friend, and I sincerly wish that you would not devote so much energy, in tilting at the wrong windmill, in the case of taking me to be an enemy. (To be clear, that's an olive branch being extended to you, however, should you desire, you may malign me as you wish, because that's certainly your perogative.)

Peace.

equus-posterior, if a few people had their lives saved on the Golden Gate but in the end wanted death, I say that is a small price to pay for those who had their life saved and figured out they wanted life. We do know that only about 6% or 30-40 of the 515 folks in Seiden's study subsequently killed themselves. That does indicate that a lot more people were happy with their saved live than were unhappy with it.

I've never said or implied that we should not provide human understanding to aid those in distress. One admittedly incomplete measure of that human understanding is the >$1 billion spent just by County Mental Health in the 30 years the barrier would be up. That measure leaves out all the churches, volunteers, hotlines, interventions by family, the Glendon Association, etc. In the pale shadow measured in dollars, probably over $10 billion is spent over 30 year son human measures. By comparision the $605,000 for the bridge barrier is very minor, and the evidence is quite strong that some lives will be saved by the bridge barrier.

I have no window into your or any anti-barrier person's soul. But a suggestion of Russian Roullette for the suicidal seems to me to be a rather base suggestion.

I do not believe that the overwhelming majority of those who are suicidal really want to commit suicide. I believe they are suffering acute distress, like a fever, and proper measures (mostly involving lots of interaction with people) can bring them to health. A suggestion like Russian Roullette for the suicidal is kind of like suggesting hallucinating feverish people be encouraged to act out strange instructions when they are not rational.

I certainly can disagree with you and dislike your examples and maintain a peaceful relationship. Thanks for the olive branch.

sevendolphins, I can agree-yet-disagree, with the idea that thoughts of suicide are an acute situation. They may be so, at a certain point, but I think that they may be part of a chronic pattern, sort of like a 'rogue' wave, in and endless set of waves. An 'attack' could occur, any time that life stresses stack up, into an overwhelming pressure-front. And then, of course, there are extreme stressors, such as the death of a loved one, loss of a job, or even rejection by a crush--and don't forget any alcohol factor, which only complicates things.

As for the Russian Roulette, that falls under the humour banner--it's not meant to be taken seriously. Sadly though, if lack of money is the supreme stressor, it might be a better option than trying to get a well-paying job, in today's marketplace. (That's not meant to be funny, either.)

Regardless, I think that even having a suicide-prevention hotline on the bridge might be a good idea, with a sign stating simply that someone cares, and try it before jumping.

To steer this back to the main issue, I still don't think it's really possible to weigh cost against effectiveness, and I'm not sure that barriers would help much. And again, while I would agree that any lives saved seems to be a good thing, I think that any money would be better spent for something that benefits the living, rather than those who seek an early end to their lives--no matter how (possibly) temporary.

If only we could spend money to mitigate the effects of the current pressures in the world to compete, and "get ahead". Perhaps, if the world were actually easier to live in each year, instead of harder, then there would be less reason for people to consider jumping from bridges.

Well, I'm deeply optimistic that people find a good, positive way through life. I think folks who go to a bridge with thoughts of suicide are indeed living people, and if they don't jump, most (if you accept Seiden, 94% or so) will go on to find their way.

A hotline with people to answer it will cost more than a barrier over the 30 year barrier lifetime. Best to build the barrier and use the difference between the hotline staffing cost and the barrier to add to the $41 million a year spent by the County on Mental Health.

I don't think Russian Roulette humor is any more appropriate than lynching humor or rape humor. All make light of a rather serious predicament, where making light of the predicament increases its acceptability.

But equus-posterior, I would suggest your views are a bit dark and negative. Eric Hoffer used to note that folks in rather bleak spots in the world (slums of Calcutta etc) are rather more happy on the average than us in the rich, industrialized world. Alan Watts used to point out that a dog or a cat tends to be pretty happy. I'd hope you can find a way to let your animal spirits buoy you, you're as valid an entity as a dog or cat or a desperately poor but happy person in Calcutta. By all logic those of us in the industrialized world should be the most able to figure out our place in the universe and be positive about it.

"I would suggest your views are a bit dark and negative."

Some, yes, but often, I'm just being something of a devil's advocate. I've known people from all ends of the personality spectrum, from those who constantly complain about easily remedied situations, to those who wear rose-colored glasses. I can still recall the words of a friend who, when asked how she was always so cheerful, stated, "I don't know, I just smile!"

Unfortunately, the human condition makes it nigh impossible, to determine how hard someone is tested in life, as each individual has different tolerances for stress. However, if one were to assume that the desire to leap from a bridge were an indication of over-stress, then I submit, that they likely have a darker view of the world around them than most of the people walking around, and perhaps understanding some of that despair would be a key to help alleviating it.

As for those in Calcutta, I would have to argue that the culture may allow for having less, equaling less life-stress, and therefore allowing more happiness. There are also buddhist monks, who go without much materialism in their lives, and are quite content. In neither place, would I expect many people to jump from bridges. However, in America, unless you're happy as a panhandler, you have to play the game, and the ante may require a mortgage and a car note.

Sorry, to stray further off-topic, but I tend to see that averages trend downward, much easier than upward. And, it's just those averages that turn people into statistics, and create a place for people to be grandly unhappy.

Overall, if you cling to "a little better, is better than none", then the barriers should be added to the bridge. Aesthetics is a non-argument, since safety should always be a larger concern than beauty, but again, I don't think throwing money at the idea of suicides is the best answer. I prefer to think that fighting suicide is about each and every individual, nurturing the connections we have to one another, and making sure that there is always someone to catch us when we stumble. (Now, you could call that overly optimistic, in counter-point to being dark, but now I'm just heading towards the other end of the line. Also, one might argue that religion could fill certain gaps in our human interactions, but I think that's a cop-out, so that we can conveniently let God(s) help to shoulder the burdens of others, when it's possible that we are failing a test put before us.)

Again, my apologies for the tangent, and another for the verbosity.

Thank you for expressing the proposition that "Aesthetics is a non-argument, since safety should always be a larger concern than beauty."

Unless notions of this sort are clearly expressed they remain among those of "our ideas we have no idea of" (Garrett Hardin's turn of phrase) that govern "thought" and action.

That the aforesaid proposition is truly preposterous is clearly illuminated by its expression.

Naturally, there will be differences of perception about this. Such are inevitable and understandable consequences of the blessings of our diversity.

"our ideas we have no idea of" (Garrett Hardin's turn of phrase)

I had never been made aware of that phrase, but have often thought a similar one: "we don't always know, what it is we don't know". . . .

Anyway, thank you, Marc, for extracting the part of my post(s), which was relevant to your article. I will try not to stray so far from the main topic in the future. (I'm a forum veteran, and should know better.)

Dr. McGinnes... sure would be nice to hear more about the sensitivity of the Glasgow approach, or why you think an internal memo saying $1,000,000 is more accurate thant he publicly stated number of $605,000.

Hi sevendolphins,

I'm happy to address your questions. Before I do, I should point out that the answers to these questions for the most part are in my report, and I've answered these very same questions previously in multiple online forums.

In response to your statement that nobody has yet devised a statistical technique that would be sensitive enough to detect a change in the suicide rate due to installing a suicide barrier, that's not quite right. The problem isn't with the techniques, but with the limited amount of data available to us -- there just aren't very many suicide prevention barriers, and they haven't been around very long. That is why I switched over to examining how *exposure to bridges* affects the suicide rate. We have a lot more data there, so the limited information problem that plagued past studies is minimized.

I assume by "sensitivity" you mean the statistical power of the test -- in contrast to past studies, this isn't really a problem with my study, as it has 1300+ observations. If there's any reasonable sized relationship between bridges and suicides, that should be enough data to detect it.

I'm not sure of the basis of your objection to the negative relationship between bridges and the suicide rate -- the data are what the data are. I explain in my report that this relationship is likely due to bridges standing in as a proxy for a strong economy or something along those lines. Further, that negative relationship disappears once I account for the unemployment rate. At any rate, there is no evidence that increased exposure to bridges increases the suicide rate.

Again, with the exception of the discussion of statistical power all of these questions are explicitly addressed in my report. And again, I have answered these questions in other online forums before before today.

Since I've been so patient answering your questions over the past month, perhaps you could answer one of mine. Who are you, and what sort of training in statistics do you have?

I was rasied in Santa Barbara and we use to party under Cold Spring Bridge for years and never had people like you have now in Santa Barbara doing the jumping. For thoses of you that are not from Santa Barbara you all need to ask a native, if they could count how many have tried back in the 60's. We use to stand on the Cold Spring Bridge on a windy day just too feel the bridge rock and the Hiway Patrol would come and run us off,so would you count that as wanting to jump...No.
Yes there are nuts out there that have tried, and still trying well! I feel let them jump if this is their choice. this Article is just like anyother article, something to talk about....and something to change. better to talk about the people you let move in to S.B.

Thanks for answering, Dr. Glasgow.

The statistical sensitivity is crucial; it determines whether or not your study could ever have detected any change from barriers. I'm not convinced your study could. Here's why.

The total number of suicides per year in the US is around 30,000, but it is not constant from year to year. It goes up and down, largely due to random factors. I see that the standard deviation in the yearly total is about 1,000, over the past 10 years, from a linear trend.

The total number of suicides by jumping from high places is about 700 per year, about 2% of all suicides. A small fraction of those suicides by falls are due to jumping off bridges, let's say, 300 per year.

How do you make any statistical case at all about those 300 suicides per year when the standard deviation on the total 1000, which is over 3 times as big *all* the bridge suicides? And barriers are an even smaller effect on the 300 bridge suicides; maybe barriers are on 1/3 of the bridges, which means you're trying to see a change of 100 when the standard deviation is 1000.

I don't see how you can ever get the sensitivity you need to make a compelling test of whether barriers influence the overall suicide rate.

You've made qualitative arguments about sensitivity with words. That doesn't convince me. Numbers are needed to back up your words.

---William Sealy Gosset aka `Student'
as for training, well, I did work with Karl Pearson, but I like Guinness better. And remember, scientific review is always anonymous.

William Sealy Gosset. Cute.

Once again, I'm *not* testing the effect of barriers on the suicide rate. I'm testing the effect of *bridges* on the suicide rate. You are correct that it would be very difficult to detect changes in the suicide rate due to barriers -- this is why I study bridges without barriers, where we have a lot more data. I explain this in my report and in my previous post to you.

I found a negative and statistically significant relationship between bridges and the suicide rate. I assume you're trying to argue that there might actually be a positive relationship between bridges and the suicide rate (which would suggest barriers can save lives), but my test isn't sensitive enough to detect it. Here are my numbers:

The regression coefficient on number of bridges had a t-ratio of 2.35, even after including the lagged suicide rate and the unemployment rate. This means that the probability that there is actually a *positive* relationship between bridges and the suicide rate (a necessary condition for suicide barriers to save lives) is 0.0095. In other words, the probability that bridges are a danger to suicidal people is tiny, and we would reject that possibility under any reasonable social science standard.

I should also note that the R-squared on this regression model was 0.84, which means the three variables in the model are explaining about 84% of the variance in the suicide rate -- hopefully this will put to rest your fears that the suicide rate is too variable over time to model. There is a lot of variation in the suicide rate to be sure, but most of it can be explained with other observed variables.

OK, I hope this helps clear things up.

To "sevendolphins":
You asked "why [I] think an internal memo saying $1,000,000 is more accurate than the publicly stated number of $605,000."

Answer: The plain fact-- whatever you or I may think bout it-- is that CalTrans has approved the sum of $1,000,000 for this proposal. I suggest that you contact the CalTrans person who is managing the CalTrans effort-- sara_von_schwind@dot.ca.gov

I sympathize with you if you are learning for the first time that some agencies, like some people, say things in public that are false. Sometimes this is intentionally deceptive, sometimes merely carelessly misleading.

In any event, I hope you will post a comment letting others know what you learn about the actual figure.

Dr. McGinnes, thanks for your comment. Of course, you were the one who represented themselves as having a deep understanding of this issue. I think that implies you know precisely how the $605,000 estimate was arrived at, and precisely how the $1,000,000 was arrived at; anything less does not seem to me to be a deep understanding.

I guess from your comment I can surmise that you really don't understand this issue particularly deeply; you are unable or unwilling to give the detailed accounting of either estimate. Instead you shirk the onus of justifying your $1,000,000 to a humble questioner, which is merely a rhetorical device. If you assert the number you should be able to deeply and convincingly defend it; it was not me, after all, who stated originally that the $1,000,000 is the best estimate for the project, it was you.

I sympathize with you if you are learning for the first time that you, like some people, go a little too far in saying things in public.

I hope you post a straightforward comment that says, `gee, I really have a shallow understanding of the cost estimates, but I chose the highest one to make the project look as expensive as I could primarily as an effort to turn people against the project for emotional, rather than logical, reasons.'

Dr. Glasgow consistently uses $605,000 as the cost estimate.

Thank you, Dr. Glasgow.

As you say, you have not directly tested the influence of barriers on the suicides. It would be very good if you stated that exceedingly clearly, perhaps in an abstract.

What you have done is attempt to take a more indirect route, but I'm about as skeptical that your indirect route holds up as you assert as you are skeptical of other studies like Seiden etc.

Is there a negative correlation between number of bridges and the suicide rate after accounting for other variables or not? Above you say:

``The regression coefficient on number of bridges had a t-ratio of 2.35, even after including the lagged suicide rate and the unemployment rate. This means that the probability that there is actually a *positive* relationship between bridges and the suicide rate (a necessary condition for suicide barriers to save lives) is 0.0095.''

In your paper (page 6) you say
``These further analyses did not find any statistically
significant relationship between the number of
bridges in a state and the suicide rate, no matter
which definition of the bridge was used.''

Which is it? It seems the first statement suggests that you still found a significant negative correlation even after accounting for other variables, while the second seems to indicate that you found no significant relationship.

In any case, whatever your residual correlation of suicide rate with bridges is after fitting confounding variables, I suggest you compute (with errors) the actual number of suicides nationally that bridges seem (by your analysis) to either prevent or induce.

And then compare that number with the *actual* number of bridge suicides. One of my concerns is that your number of bridge suicides deduced by correlation analysis greatly exceeds the actual number of bridge suicides, and that also the statistical error on your number of bridge suicides substantially exceeds the number of bridge suicides.

Oh, I see, sevendolphins. You just want to vent your spleen regardless of the facts and without bothering to use the information I gave you so you could contact CalTrans to verify the $1,000,000 figure.

I am not sure why this is, but you can be sure that I have no further time or attention for you on this point.

Thank you, Dr. McGinnes. Actually, I was doing my best to ascertain from you, since you professed a deep undersanding of the issue and put forward a cost of $1,000,000, how you arrived at that figure. Dr. Glasgow uses $605,000, which is also the figure on publicly available documents on the Caltrans website.

Thanks for letting me know that you do not actually have any understanding of the difference between the $1,000,000 and $605,000. I sympathize with your frustration that a humble questioner discovered this fact.

Dr. McGinnes,

Thanks for the e-mail of Sara von Schwind, her
reply is that the principal difference between the
$605,000 and $1,000,000 is potential cost of a
more aesthetically pleasing design. She says
possible mitigation and permitting costs also make
up the difference.

Of course, if the project costs even more than
$1,000,000, that would not be the first time our
government or Caltrans (or your prior place of
employ, UCSB) spent more than their
initial estimates. It is not really necessary to assert
that anyone is making false statements. More like
the usual unforeseen events of fate and randomness.

>>"As you say, you have not directly tested the influence of barriers on the suicides. It would be very good if you stated that exceedingly clearly, perhaps in an abstract."

My report first reviews past research which has tested for the effect of barriers, none of which found a statistically significant effect. I then conduct a new statistical study looking at the effect of bridges on the suicide rate, and again found nothing statistically significant. I think you're reading through my report too fast ...

>>"Is there a negative correlation between number of bridges and the suicide rate after accounting for other variables or not?"

There is with a simple regression. There is not with a fixed-effects regression with heteroskedasticity-corrected standard errors. In any case, I could not find a positive relationship, which is what we'd expect if suicide barriers save lives.

>>"In any case, whatever your residual correlation of suicide rate with bridges is after fitting confounding variables, I suggest you compute (with errors) the actual number of suicides nationally that bridges seem (by your analysis) to either prevent or induce.

And then compare that number with the *actual* number of bridge suicides. One of my concerns is that your number of bridge suicides deduced by correlation analysis greatly exceeds the actual number of bridge suicides, and that also the statistical error on your number of bridge suicides substantially exceeds the number of bridge suicides."

I should first note that your concern actually isn't a problem with regression analysis. The regression line is computed using the actual data, so there is no way predictions using this line can greatly exceed the actual number of bridge suicides (as a quick look at the scatterplot in my report will reveal). The only time you need to worry about the regression line and the actual data diverging is if you are trying to make forecasts outside the bounds of the observed data (which I am not doing here). Nevertheless ...

The coefficient on bridges in the regression I mentioned in my previous post was -0.02. The would come out to one fewer suicide per 100,000 people for every 50 bridges over 30m high in a state. The most bridges over 30m observed in a state in any given year was 59. The predicted change in the suicide rate in going from 0 to 59 bridges is 1.18 per 100,000, with a standard deviation of 0.56. The actual suicide rate per 100,000 per state-year varies between 4 and 28.8, and the actual suicide rate by jumping per state-year varies between 0 and 2.2.

Again, think of regression as a trend line fit to the actual data -- predictions using the regression line will never diverge from the data unless you are making predictions outside the bounds of the observed data (such as predicting the suicide rate if a state were to build 1000 tall bridges).

Hopefully this alleviates your concerns on this point.

Thanks Dr. Glasgow.

>>>"Is there a negative correlation between number of bridges and the suicide rate after accounting for other variables or not?"

>>"There is with a simple regression. There is not with a fixed-effects regression with heteroskedasticity-corrected standard errors. In any case, I could not find a positive relationship, which is what we'd expect if suicide barriers save lives"

Recall you said:

``The regression coefficient on number of bridges had a t-ratio of 2.35, <b>even after including the lagged suicide rate and the unemployment rate.</b> This means that the probability that there is actually a *positive* relationship between bridges and the suicide rate (a necessary condition for suicide barriers to save lives) is 0.0095.''

Where I've made tried to make a phrase bold (can't tell if it worked with the preview). Can you explain the conditions under which you found a t-ratio of 2.35, and a probability of a positive relationship between bridges and the suicide rate of 0.0095? Did you get that t-ratio of 2.35 and probability of 0.0095 for a a fixed-effects regression with heteroskedasticity-corrected standard errors? If not, what was the t-ratio and probability for the
fixed-effects regression with heteroskedasticity-corrected standard errors?

I'm finding I can't follow your answers, perhaps because of my own ineptitude, so I'll focus on one item at a time. I have several more.

Hi again,

My workload is picking up at UCSB, so I might not be able to check in here regularly. Nevertheless, I'll try to answer as many questions as I can.

Incidentally, a fixed-effects regression with heteroskedastic standard errors is not that much different from an ordinary regression -- it accounts for the different baseline suicide rates in each state, and corrects for smaller states having more variation in the suicide rates over the years (due to their smaller population). However, the basic image to keep in mind is a line running through a scatterplot that describes the trend in the data, just like an ordinary regression.

At any rate, the t-ratio for that model is -1.69, which yields a probability of a positive relationship between bridges and suicides of 0.0456.

Thus, there is a positive probability that bridges do cause suicides (as there always must be with any statistical model), but the probability is too low to place much faith in that assertion.

Thanks, Dr. Glasgow.

What are the differences between the regression that
gave a t-ratio of 2.35 and consquently a probability of a positive relationship of 0.0095 and that that gave a t-ratio of -1.69 and consequently a probability of a positive relationship of 0.0456? Why is the first t-ratio (2.35) positive, and the second (-1.69) negative?

Hi again,

Sorry, I was being sloppy there. Both t-ratios are negative (-2.35 and -1.69) -- I accidentally omitted the negative sign in my earlier post. Any negatively sloping regression line will have a negative t-ratio, and any positively sloping line will have a positive t-ratio.

T-ratios are calculated by simply dividing the estimated regression slope by the estimated standard error. This gives you a measure of the distance from zero for your estimated regression slope. Larger t-ratios mean your estimate is more standard errors away from zero, and thus it is less likely that the real relationship is zero (or the opposite sign). If the t-ratio is large enough, we call our estimate "statistically significant" -- this means the probability that the real relationship is zero is low enough that we choose to reject that possibility and conclude the real relationship must be in the same direction as our estimated regression slope.

As far as the differences in the regression, the regression that produced a t-ratio of -2.35 was just a simple regression -- basically just fitting a line to the scatterplot in the figure in my report. The t-ratio indicates that we can be reasonably sure that there actually is a negative slope on the line in this model. The model that produced a t-ratio of -1.69 was a fixed-effects regression with heteroskedastic standard errors -- it corrects for the different baseline suicide rates in each state, and it also modifies how it calculates the standard errors (you need to correct for the different variances in the suicide rates across states) -- these two things together will change the regression slope and standard error, which changes the t-ratio.

In both cases, these models produced negative regression lines and somewhat large t-ratios, which gives us a low probability that the real relationship is positive.

Sorry again for any confusion caused by my typo.

Thank you Dr. Glasgow.

I'm still confused, probably due to my own ignorance.

For the simple regression depicted on page 6 of your paper, the regression line describes about -5 /100,000/30 bridges, or coefficient of -0.166 (per 100,000 per bridge). With a t-ratio of -2.35, one would then say the result of the coefficient is:

(-0.166 +-0.071) suicides/100,000/bridge.

Then, when you the more involved regression with
heteroskedastic errors and (I think) allowing for
the unemployment rate correlation, you get a coefficient of -0.02 with a t-ratio of -1.69. So, that looks to me like your coefficient is:

(-0.02 +-0.012) suicides/100,000/bridge.

The standard error has fallen from 0.071 to 0.012... by a factor of 6. Do you know why that is?

I think you have added a new variable to the regression, which should have *increased* the error. However, perhaps accounting for the heteroskedasticity reduced the standard error. Hard to say without knowing the details.

From WISQARS, the total suicide rate from falls in 2004 was 0.23 per 100,000. Say 1/2 of those are from bridges, so, 0.12 per 100,000.

Say that the mean number of bridges in a state is 3. That's a guess from your figure on page 6.

Your plain regression would then say bridges appear
to save 3*(0.17+-0.07)=0.5+-0.2 per 100,000.
Since 0.12 per 100,000 actually do jump from bridges, the result appears to indicate that for every true bridge jumper, there are (5.2+-1.7)/100,000 folks who don't commit suicide because of the presence of bridges. But that does not account for the confounding effect of other variables like unemployment.

The second regression that accounts for unemployment and uses heteroskedastic errors. That one would indicate that for every true bridge jumper, there are (1.5+-0.3)/100,000 folks who don't commit suicide due to the presence of bridges.

BTW, I don't think the proper baseline is 0... we *know* that some people do jump off bridges. To see your negative effect, you first have to cancel out the known bridge jumpers, which I've assumed to be 0.12 per 100,000.

But I am suspicious of one of the standard errors in the (0.17+-0.07)/100,000/bridge and (0.02+-0.012)/100,000/bridge above. Adding one variable and accounting for heteroskedasticity don't seem to me likely to account for an error reduction of a factor of 6.

If the first regression has a correct error, my point is that even the error excedes the true number of bridge suicides per 100,000.

If the second regression has the correct error, there is still a major surprise: that for every bridge jumper 1.5 folks are induced somehow to not jump due to the presence of bridges.

Isn't it more likely that there are further confounding variables that are inducing a spurious negative correlation?

Note these regressions are for the *total* suicide rate, not the suicide rate *by jumping*. Since jumpers are a small fraction of total suicides, it isn't surprising that the standard error on the coefficient on the number of bridges would be larger than the total number of bridge suicides. I did find a positive relationship between the suicide rate by jumping and the number of bridges, as we would expect.

The negative relationship between the total suicide rate and the number of bridges is likely spurious, as you note and as I acknowledge in my report. I believe that bridges as acting as a proxy for urbanization and/or a strong state economy. It could be that the actual relationship is positive, but I've been unable to detect this positive relationship trough several different types of regression, several different types of definitions of bridges, and several different permutations of control variables. Thus, I concluded that I was unable to find evidence that bridges cause suicides.

It all boils down to this: if suicide barriers save lives, we would expect to see a positive relationship between bridges and the suicide rate. However, I was unable to find such a positive relationship no matter what type of regression or what control variables I included. Note I'm not saying I've proven barriers *don't* save lives -- only that I and everyone else that has studied the problem has been unable to prove that the *do* save lives.

Thanks, Dr. Glasgow. I am a terrible communicator, so let me try again to pose a question because my first one must have been unclear:

Why is the standard error on the coefficient obtained with unemployment rate and the heteroskedasticity:

(-0.02 +-0.012) suicides/100,000/bridge

six times smaller than the error on the coefficient obtained with the simple regression

(-0.166 +-0.071) suicides/100,000/bridge ?

Thanks for your lengthy answer on other topics. Isn't the situation like a person with uncorrected 20/200 vision being unable to read the letters that test for 20/20 vision?

Just because they can't see the letters doesn't prove the letters are absent, and in fact, their failure to see them isn't even interesting, since their vision is so blurry they couldn't see them in any case. However, a useful experiment is to get someone with known 20/20 vision to look for the letters.

Your method with the available data sample does not have sufficient sensitivity to make any decisive conclusion on the influence of bridges on total suicide rates, because your errors (both systematic and statistical) substantially exceed the known suicide rate from bridges.

Sorry to be away for a while.

I think I see what you're trying to say -- that the negatively sloping regression line that describes the relationship between bridges and all types of suicides does not rule out the possibility that bridges do cause at least a few suicides, since bridge suicides are relatively rare. This is true.

The proper conclusion to draw from my study is that I was unable to find evidence that bridges do cause suicides, and thus evidence that suicide barriers save lives. You're right that this doesn't rule out the possibility that barriers *might* save lives -- but we have no evidence of this.

Note that this isn't a problem unique to my study or statistical method. *Nobody* has been able to find evidence that suicide barriers save lives. It could be that this is because the data are too noisy to pick out the lifesaving effects of barriers, or it could be because barriers just don't work. We have no idea which is the case.

Thanks, Dr. Glasgow. I'm still very curious about the answer to the question:

Why is the standard error on the coefficient obtained with unemployment rate and the heteroskedasticity:

(-0.02 +-0.012) suicides/100,000/bridge

six times smaller than the error on the coefficient obtained with the simple regression

(-0.166 +-0.071) suicides/100,000/bridge ?

I cannot understand without some further explanation how changing the conditions of the regression would cause such a large change in the standard error on the coefficient.

When you say `that the negatively sloping regression line that describes the relationship between bridges and all types of suicides does not rule out the possibility that bridges do cause at least a few suicides'

There is a key point that I'm not sure I've communicated well: it is that the `few suicides' are **all of the suicides that occur off of bridges**. The regression that you describe w/respect to the *total suicide rate* is not sensitive enough to find the *known signal* of bridge jumpers. Which is not to say that you *had* to find that signal... you might have found substitution or no signal at all, but the point is such a finding only matters when your errors are small compared to the known rate.

You did show that if you limit yourself to a regression of *known bridge suicides* versus number of bridges, you did see a positive effect. The correlation coefficient you got with that regression was well beneath your sensitivity in the regression w/r to the *total* suicide rate.

As to other studies, I think you argue that *every single one* of the 515 folks interviewed by Sieden was irrelevant. I don't accept that... I think the flaws you identified in the Seiden study reduce its significance, but do not eliminate the significance entirely.

The standard error is smaller in the 2nd regression because regression without correcting for heteroskedasticity is inefficient -- the standard error estimates will be biased upwards. The heteroskedasticity correction solves this problem.

I understand what you're saying, and I agree with you -- an increase in jumping suicides due to bridges might be too rare to detect in my statistical study, even though I've expanded my data far beyond what researchers have looked at in the past. This could be. Again, my study is not proof that barriers are ineffective. Instead, my study joins a long list of studies that were unable to find any evidence that barriers are effective.

One other point to consider -- some studies of barriers have been conducted in cities where the proportion of suicides by jumping is quite high. For instance, in Bern jumping constituted about 30% of all suicides in the city. After a suicide barrier was installed at the most popular jumping spot, suicides by jumping fell -- but there was no significant change in the overall suicide rate. Thus, even in situations where suicide by jumping is relatively common we cant find any evidence that barriers save lives.

The major problem with the Seiden study is that it doesn't actually address the issues of displacement and substitution. The easiest way to see this is through a thought experiment -- imagine we could go back in time and install a suicide prevention barrier on the Golden Gate Bridge when it was built. What would happen with the 515 people in Seiden's study? Would they still go to the Golden Gate Bridge even though there is a barrier there, or would they go somewhere else? If we want to argue that the Seiden study is proof that barriers save lives, then we must assume these people would have gone to the Golden Gate Bridge even though there was a barrier there -- in other words, we must assume there is no displacement or substitution.

Thanks, Dr. Glasgow.

I've not seen the improvement in statistical efficiency due to heteroskedasticity ever be a factor of 6; at most a factor of 2. Can you check that the coefficient with heteroskedastic errors was really -0.02 and not -0.2?

I'm still not communicating the following points: I understand that your study did not find any correlation of total suicide rates with # of bridges, and thus could not deduce any potential benefit of barriers. My point is that your errors are so large that your inability to find a correlation does not matter; a study like yours with the data sample available has insufficient sensitivity to be useful for any decision making.

As to Bern, the words you use are interesting but not decisive. The decisive issueis whether the error on the overall suicide rate was small enough to discern the change that could have been expected from the total absence of the bridge suicides. If the error is not small enough, no statistically significant conclusion can be drawn. You did not quote the errors and values of the rates, so I can't tell.

I think some of the 515 Seiden folks would have gone to the Golden Gate Bridge even if a barrier was there. I don't think all of them would have gone, but some of them would have. Very distraught people don't double check such issues before they act. I remember one story, I think a GGB survivor, who decided to walk to the bridge, and then if a single person smiled back at him during the walk, they'd not jump. No-one smiled back. They jumped. Maybe I've garbled the story, but my point is some people are very distraught and not fully rational when they decide to seek a place to commit suicide.

Seiden's result does not dependon assuming *no* displacement, you are exaggerating there. It depends on whether displacement vitiates his result entirely. I don't think displacement vitiates his result, although it may reduce its significance.

Dr. Glasgow... a fruitful idea was mentioned in another blog. It is: simply poll people in communities where their are barriers on bridges and ascertain what fraction are aware of the barriers. I doubt such a study will find that 100% of people in a community are aware of the barriers. Heck, only 75% of people know that the earth orbits the sun, according a recent survey.

I think you are arguing that 100% of potential suicides are aware of the barriers on the bridges in their communities or whereever they plan to jump. I don't think that 100% is a reasonable assumption.

The conversation has moved over to blogabarbara, where new lows in civility and reading comprehension have been achieved.

What is blogabarbara?

http://blogabarbara.blogspot.com/

But I think you already knew that ...

Thanks Dr. Glasgow. No, I did not know about that website. If you prefer, I'll post there.

And stopping one war somewhere is useless as well because another war will just break out somewhere else.

Astronomically dunderheaded.