Hold ’em Harmless?


The following letter was sent by US mail on 26 October 2015 to the Delhi District office of State Senator John J. Bonacic.  An electronic version  with attachment was transmitted to the  Senator’s  e-mail three days later.



Dear Senator Bonacic,

At the hearing you held Sept 9 regarding S5302 there was good news: you gave at least a little time to the important question of whether legalizing i-poker would have an impact on problem gambling and gambling addiction. The bad news was that you readily accepted a “negdec”   from Mr Pappas of the Poker Players Alliance. I fear your questioning was to get this assurance of no harm onto the record.

Your questioning of Mr Pappas did not show the trial lawyer skills that Mr Featherstonaugh accorded you later in the hearing. Was this just a lapse in preparation, or was it deliberate? Whichever it was, your “OK” to Mr Pappas’s reply surely gave most listeners the false impression that internet gambling — of all kinds – has been well-studied and found not to be worse in any dimension for individuals or populations than other kinds of gambling. Not so.

I would be glad to meet with you and your staff to go over some basic principles of epidemiology and public health that should be applied to the important work you and your colleagues do. They are explained in the enclosed 12-page critique. Sad to say, the approach in the September hearing to this basic science  is no more valid than evaluating a corporation by whether it declared a profit or loss in the most recent annual report.


Stephen Q. Shafer MD MPH

Chairperson,  Coalition Against Gambling in New York  917 453 7371

Below is the critique that was enclosed with the cover letter

Considerations of Internet Problem Gambling in the New York State Senate i-Poker Hearings of September 9 2015: an Epidemiologist’s Critique

Stephen Q. Shafer MD MA MPH                                              27 October 2015


The author is a retired Clinical Professor of Neurology, Columbia University and Chairperson of Coalition Against Gambling in New York, a non-profit all-volunteer organization registered in Buffalo.


During the Sept 9 2015 hearing on legalizing i-poker held by Senator Bonacic there was scant mention of the potential for i-poker or other forms of i-gambling to cause addiction or problem gambling or to sustain these conditions when they had developed in another setting such as a b and m [bricks and mortar] casino. Below is the nearest approach.

At about 15:50 Mr Bonacic, chairing, asked Mr Pappas, CEO of the Poker Players Alliance and the first person to testify, “Is there a ratio for the amount of people that play on line poker, gaming, as opposed to those that get addicted? Is it one in three hundred, one in five hundred?   Is it ascertainable?”

Senator Bonacic seems here to be groping for the prevalence of gambling addiction among persons who do i-poker or i-“gaming.” I expect he meant to make the ratio as he set it up  500 to one, not one in 500. He is certainly leading the witness towards a very low proportion of addicted gamblers among all on-line gamblers.  Note also that the question does not separate poker from other types of i-gambling. This is likely intentional, to blur distinctions in readiness for the transition I think he and associates plan, from two particular forms of i-poker to all forms of casino-type “gaming” on the internet and ultimately to i-betting on sports.

 Mr Pappas responded that he didn’t have notes in hand but that his written testimony gave backup. He summarized,    “There is not a discernible increase, not any increase.” Mr. Bonacic replied “OK. ”

Note well: Mr Pappas did not answer the question Senator Bonacic asked, which was about a “ratio,” not an increase. His response belonged to a question not put. Perhaps he had been expecting something like one of the following:

  1. Is someone who plays i-poker more likely to become a problem gambler than someone who plays live at a casino or card room?
  2. Is someone who plays i-poker more likely to become a gambling addict than someone who plays live at a casino or card room?
  3. Is someone who gambles on the internet with poker and casino-type “games” more likely to become a problem gambler than someone who does these live at a casino?
  4. Is someone who gambles on the internet with poker and casino-type “games” more likely to become a gambling addict than someone who does these live at a casino?
  5. Does the availability of i-poker increase the prevalence of problem gambling in the population?
  6. Does the availability of any and all forms of internet “gaming” increase the prevalence of problem gambling in the population?

No matter what Senator Bonacic meant to bring out (or to obfuscate) with his query, most hearers of the dialogue will interpret the response, with his OK, to mean that legalizing gambling on the internet will not increase the number of problem gamblers in the population . That is not a valid conclusion from the references Mr Pappas cites in writing or from any other compilation of references .  This is a complex research question. It has not been answered yet and may never be unless all privacy on the internet is abolished.

Mr Pappas, in his written testimony (with its references) and oral responses , provides two lines of reasoning purporting to show that internet gambling is not a risk factor for problem gambling or gambling addiction.  One comes in his [I think deliberately] tangential response to Mr Bonacic’s question: looking at time trends in prevalence statistics.   The other, he did not take up in speaking, though it gets more than a few pages in the 610 page compilation of papers on gambling available at the web site of the PPA. This relates to the effect of i-gambling on the individual, one determinant of impact on the population, but not the only one. It seems Mr Pappas intends the reassuring phrase “not any increase” to apply as the hearer wishes to population frequencies or individual risks. There is no evidence it fits either. I’ll look at population effects first, then individual risks.

Mr Pappas’s testimony reads

“… a key report on American online gamblers last year from the nearby University at Buffalo Research Institute on Addictions – proves [emphasis added] that online gaming does not increase the social risks and damage of problem gaming6.”

The report cited, by Welte, Barnes, Tidwell et al published in J Gambling Studies in 2014, proves nothing of the kind.    This follow-up to the earlier national survey by the same highly respected group showed that between 1999 and 2013, years in which the internet became more widely available, there was no significant change in the prevalence of problem gambling nationally although the proportion of the population who gambled on the internet rose from 0.3 % to 2.1%.  This jump, though relatively large (a sevenfold increase) does not permit any conclusion about an important question, number 6 in the list above . Does the availability of any and all forms of internet “gaming” increase the prevalence of problem gambling in the population?

This question could theoretically be answered by a population-based study like those the RIA has done so well, but not by that particular study even with its relatively large sample size, close to 3000. The proportion of the population newly exposed to the putative risk factor (internet gambling) is very low. Also, the precision of the prevalence estimate does not allow detection of small changes in prevalence, however important they might be on a national scale.   A detailed explanation of why Mr Pappas’s conclusion of no increased risk to society is invalid is in the appendix.

If trends in the best population-based study to date don’t point either way about whether i-gambling is more addictive than other forms of gambling, can something be learned from the experiences of individual i-gamblers?   The PPA compilation of studies contains nine articles from the Division on Addictions of the Cambridge Health Alliance ( hereafter DOACHA), a prolific research group headed now for many years by Howard J. Shaffer, Ph.D. Much of the group’s work regarding i-gambling concerns whether analyzing on-line gambling behavior can detect problem gambling well enough to guide regulatory policy and aid preventive efforts .   There are hints in these papers, but no definitive results.   Significantly , only one paper relates to i-poker alone.  Most report on sports betting and to a lesser extent i-casino. Both of these were in the study years much more popular than poker among subscribers to the gambling ISP bwin.party, which allowed the researchers to approach its subscribers on strict conditions.

The DOACHA research found, to no one’s surprise, that a relatively small number of participants bet more, lose more and are more “ involved” than the vast majority. Privacy concerns, however, make it impossible to tell for sure who among them is a problem gambler even when subject takes a suggestive action like closing his account.

In the best effort to date to learn the prevalence of problem gambling among internet gamblers (recall Senator Bonacic’s question to Mr Pappas on Sept 9), researchers from the Division on Addictions used their longstanding cooperation with bwin.party to administer a three-question survey (BBGS) to 1440 subscribers who had been using bwin.play for at least a year. None played poker. This group comprised 1.4% of the 100,000 subscribers who had been invited to take a brief confidential survey. 27% endorsed at least one of the three items. This “BBGS +” subgroup played more different “games” and more often than the other 73%.  Drawbacks to the study besides the low sampling rate were (a) that the screening instrument, the BBGS, was not validated and (b) there was no way of knowing if the respondents also used other gambling platforms on or off the internet and what behaviors they showed thereon. Thus it was not possible to estimate the actual frequency of problem gambling or disordered gambling even in the final sample, still less in the universe of bwin.play users. Reference LaPlante DA. Nelson SE Gray HM Breadth and Depth Involvement: Understanding Internet Gambling Involvement and Its Relationship to Gambling Problems. Psychology of Addictive Behaviors 2014 28:2, 396-403.

Tom and colleagues at DOACHA detailed more findings, showing that in all the games a small fraction of participants lost the most. Tom MA, LaPlante DA, Shaffer HJ Does Pareto Rule Internet Gambling? Problems Among the “Vital Few” and “Trivial Many” Journal of Gambling Business and Economics 2014 8:1 73-100   For the hold (net losses of all users)   on fixed odds sports betting for example,   99 accounts (“the vital few”) provided 80% of the hold. Forty-seven of the 99 (53%) were BBGS +. Of the 1262 accounts (“the trivial many”) that left the other 20% of the hold, only 26% were BBGS +. Neither of these papers purports to estimate the frequency of problem gambling among established bwin subscribers. Even if this estimate could be made, it would have to be compared to one for gamblers who do not use the internet to tell whether the internet increases risk.

The paragraph below in italics, excerpted from the massive compilation by the PPA, glosses over too many uncertainties to be credible.

Report for the Massachusetts Treasurer’s Online Products Task Force prepared by Spectrum Gaming Group p. 194 [found in the PPA compilation]

 On March 22. 2012, the Task Force met with Kathleen Scanlan and Jum Wuelfing of the Massachusetts Council, where findings of from [sic] the bwin,party DOA collaborative were presented. Among the findings from the bwin,party DOA collaborative

  • Problem gambling rates on the internet are not significantly different from problem gambling rates observed with other forms of land-based gambling.
  • Problem gambling indicators are less associated with magnitude of betting or volume or transaction but more indiscriminate betting across multiple and diverse products.
  • Self-imposed limits are a stronger identifier of problem gambling that site-imposed limits
  • Tracking software and data analytics can be used to identify potential problem gamblers early on for remedial action

It implies, for example, that “problem gambling indicators” have been validated well enough to allow accurate estimates of prevalence of problem gambling. This has not happened with enough precision to make them useful for prevention and treatment, only as topics for further research, which will always be inconclusive. Why?

“Responsible” gambling operators cannot afford to really recognize most problem gamblers. If they could, then according to their “ethic” the operators would have to try in good faith to see that those problem gamblers are guided to recovery.   Such a good faith effort is a conflict of interest for casinos, which depend on problem gamblers for 50% of their “gaming revenue.”

Casinos are good at spotting the gamblers who are making visible problems for management and showing them the door.  See the discussion I transcribed between Ms Jennifer Shatley, of Caesars, and Gaming Commissioner Sample at the April 9 2015 Gaming Commission Forum on Problem Gambling . Casinos must , however, blind themselves to the vast majority of problem gamblers, who are hurting themselves and their families yet not making scenes on the “gaming” floor.

Internet gambling operators are no different from casinos. They proudly fund and cite research which always concludes that problem gamblers do exist but cannot be reliably detected at an early stage without invading privacy.

The fourth bullet point above is literally correct in saying “can be used” but all wrong about efficacy. A squirt gun “can be used” to fight a house fire.

In i-poker a customer’s history is known so well that a computer algorithm could put up red flags.    Mr Pappas of the PPA replied to a question that i-based gambling is therefore better set up to detect actual or incipient problem gambling than are bricks and mortar casinos. He noted that b and m facilities cannot track a user through live table games or live card games. [ “Games “ using loyalty cards, though, can be tracked and losses tabulated. ]   Several papers from the Division on Addictions demonstrate that some markers of possible problems can be picked up in a user’s  history of on-line “play.” They have not yet been validated for clinical utility nor will they ever be.

Terabytes of data and months of 24/7 discriminant function analysis will not provide reliable markers of problem gambling unless an unbiased sample of i-gamblers is interviewed in depth by skilled interviewers or at least respond fully to an online questionnaire far more penetrating than the BBGS. This is very unlikely ever to happen. So, the sensitivity, specificity and the positive predictive value, the   operating characteristics of any proposed marker, cannot be determined.

A very thorough computer search could be done without the subject’s consent, going well beyond gambling behavior. This is off limits  if it goes beyond getting a credit history .

Would an ISP be willing now to send a message of concern to a subscriber who has screened positive with a candidate marker for possible problem gambling? Maybe, maybe not. Who is ever to know?

Mr Pappas overlooked, by the way, that in internet gambling personal appearance, behavior, speech, and body language are invisible.  Bricks and mortar casinos can theoretically perceive these behaviors (though they seldom try hard). There are no tipoffs on the internet about income [unless the subscriber has asked for deposit limit higher than the norm and volunteered evidence ]. At land-based casinos staff could easily learn what kind of car a customer has or have someone scope out the person’s house from the sidewalk [though they would never admit to doing so].  Casinos also have a huge data base on electronic device customers (the majority) through loyalty cards .

In internet gambling there are also no ugly drunks accosting other users for loans or screaming about being cheated or kicking machines, behaviors that DO trigger alarms for casino staff. American casinos have a lamentable record of spotting and helping problem gamblers who are not thus making nuisances of themselves. They defend themselves by saying they had no idea. The Ameristar lawsuit is a good example.   How can we expect a computer to do better?



Appendix Why a well-done study of the population prevalence of problem gambling cannot be expected to show the effects of i-gambling.

I will not apologize for using notation and terms not familiar to everyone. They are unavoidable. To discuss public health and individual risks of disease without certain terms is like analyzing a company’s performance entirely by profits  [or losses]. If you know the terms relative risk, conditional probability, and prevalence the next section will be an easy review. If you don’t, I hope they will become clear.

How might legalizing i-gambling affect the distribution of gambling behaviors in the population?   Increasing the availability of i-gambling can change the proportion of persons (prevalence) in a population who are problem gamblers in two possible ways or in an overlap of the two. One is to convert non-gamblers to gamblers, with a small fraction of the new gamblers becoming problem gamblers. The other is to make users into problem gamblers more often than occurs with non-internet gambling. This possible effect is expressed as relative risk or risk ratio, abbreviated RR.

In general terms, RR compares one probability to another. It is usually the ratio of two conditional probabilities, each expressed as the probability of a certain state or event depending on some condition such as exposure. P(PG) represents the unconditional probability that someone is a Problem Gambler if all we know about him or her is membership in a defined population, say of all adults in America. A condition modifies a probability, pinning it to a subset of a larger population.   It appears to the right of a perpendicular line symbol | which stands for the word “given.”

For example,

P(PG|Gnet) is the proportion of internet gamblers who are problem gamblers. Likewise, P(PG|Gnotnet) is the proportion of gamblers that do not gamble on the internet who are problem gamblers.

Necessary  assumption: all gamblers are in one of the two groups: Gnet or Gnotnet

The ratio of P(PG|Gnet) to P(PG|Gnotnet) is the risk ratio for being a problem gambler associated with gambling on the internet vs. gambling only elsewhere. This relative risk or risk ratio (RR) is often written as P(PG|Gnet) divided by   P(PG|Gnotnet)   or simply P(PG|Gnet) / P(PG|Gnotnet).   Shorthand can be RRnet   A risk ratio of 1 indicates no effect, no difference between the two conditional probabilities.

Note: Ideally, risk ratio is estimated from new cases per time, true incidence rates. When incidence cannot be known, prevalence has to stand in.

Consider now a population of 1,000,000 adults, 700,000 of whom gamble at least sometimes (G) while 300,000 (notG) never do. The probability of G is estimated by the proportion of gamblers in the population = 700,000/1,000,000 = 0.7  The probability of notG is 300,000/1,000,000 = 0.3

P(G) + P(notG) = 1 because all persons are in one of the two subgroups.

Now split the 700, 000 persons in G into two groups.

3,000 gamble on the internet; thus P(Gnet) = 3000/1,000,000 = 0.003

697,000 gamble elsewhere; thus   P(Gnotnet) = 697,000/1,000,000 = 0.697

Baseline case: choose an arbitrary but realistic value of 0.057 for probability of being a Problem Gambler given being a net gambler P(PG|Gnet) and   0.057 for probability of being a Problem Gambler given being someone who gambles, but not on the internet P(PG|Gnotnet).

P(PG|Gnet) / P(PG|Gnotnet) = (0.057 / 0.057) = 1 In shorthand, RRnet is 1.

In this hypothetical scenario, among all gamblers, use or non-use of the ’net has no effect on the probability of being a problem gambler. Promoters of internet gambling want legislators to believe this is the case. Really, though, no one knows.

Let the abbreviation  # Gnet in a population of 1,000,000 = P(Gnet) * 1,000,000 = .003 * 1,000,000 = 3000. Same idea for the abbreviations # Gnotnet and # notG.

The number of Problem Gamblers in the entire population is given in the next statement,

P(PG|Gnet) * #Gnet   +   P(PG|Gnotnet)* #Gnotnet   + P(PG|notG) * # notG. Then,

.057 * 3000                 +      .057 * 697000                    + 0 * 300,000  = 171 + 39729 + 0 =39,900

Therefore P(PG) is 39900/1,000,000 = .0399 = 3.99% This is the prevalence of problem gambling in the entire hypothetical population of one million adults. In the table below, these numbers are in the row called baseline # PG:   171 in col A; 39729 (= .057 * 697000) in col B; and zero in col C. (One cannot be a problem gambler without being a gambler.)

Let’s move from baseline to a later year, called time 1. By now, 2.1% of the population , or 21,000 persons,   gamble on the internet. There are 18,000 more net-gamblers in column A than there were at baseline,  but we will assume in this first post-baseline scenario that all 18,000 came from column B, the sector who gamble but not on the internet. Assume also that internet gambling does not lead to problem gambling any more than does gambling other ways. That is, RRnet = 1. In this case, time 1 scenario A, there is no increase in the population prevalence of Problem Gambling from baseline. It is still 3.99% though relatively far more persons are i-gambling than at baseline.

.057 * 21,000             +      .057 *679,000          + 0* 300,000  = 1197 + 38703 + 0 = 39,900

39,900 / 1,000,000 = 3.99%


scenario Gamble on Gamble not not gamble sum cols pop
net only on ‘net A thru C preval.
baseline # persons 3000 697000 300000 1000000
baseline # PG 171 39729 0 39900 3.99%
time 1 A # persons 21000 679000 300000 1000000
time 1 A # PG 1197 38703 0 39900 3.99%
time 1 B # persons 21000 687000 292000 1000000
time 1 B # PG 1197 39159 0 40356 4.04%
time 1 C # persons 21000 679000 300000 1000000
time 1 C # PG 2394 38703 0 41097 4.11%
time 1 D # persons 21000 687000 292000 1000000
time 1 D # PG 2394 39159 0 41553 4.16%
time 2 E # persons 100000 600000 300000 1000000
time 2 E # PG 5700 34200 0 39900 3.99%
time 2 F # persons 100000 647000 253000 1000000
time 2 F # PG 5700 36879 0 42579 4.26%
time 2 G # persons 100000 600000 300000
time 2 G # PG 11400 34200 0 45600 4.56%
time 2 H # persons 100000 647000 253000
time 2 H # PG 11400 36879 0 48279 4.83%

In time 1 scenario B we assume RRnet = 1 but now assume that of the 18,000 individuals added to column A since baseline, 10,000 came from column B and 8,000 from col C , the non-gamblers. The number of problem gamblers is now calculated like this:

.057 * 21,000   +            .057 *687,000                  + 0* .292  = 1197 + 39159 + 0 = 40,356 .

Thus the population prevalence of problem gambling is   40,356 / 1,000,000 = 4.04%.

This figure is higher than 3.99% .

A sample with 3,000 respondents would not find the difference between 3.99 and 4.04 “statistically significant.”   In a national population of 230,000,000 adults, however, each increase of one one-hundredth of one percent in the prevalence of problem gambling means the addition of 23,000 more problem gamblers to the population. Differences too small to be detected even in a substantial sample can mean a hundred thousand new cases nationally.

Time 1 scenario C varies from scenarios A and B in assuming that RRnet = 2 , that gambling on the net doubles the probability of becoming a problem gambler but that internet gambling recruits only from column B, other types of gamblers.   By the same arithmetic as above, we see that the prevalence of problem gambling is 4.11%

In time 1 scenario D, RRnet = 2 and the 18,000 additions to the ‘net gambling sector include 8,000 persons converted from not-gambling to internet gambling. The prevalence is now up to 4.16%. This is still much too small for a sample of 3000 to have a reasonable chance of discovering a “statistically significant” difference. To have that reasonable chance, a sample of over 200,000 would be needed. Yet if the true prevalence of problem gambling went from 3.99% to 4.16% in a population of 230 million adults, the number of problem gamblers in the nation would be increased by 391,000.

Scenarios E, F, G and H repeat the same sequence as A-D except that we are now at time 2, when 10% of the population gamble on the internet. From E it is evident that if RRnet = 1 and no one is recruited to ’net gambling who used not to gamble, prevalence of Problem Gambling does not change no matter what proportion of the population gamble on the internet.   If, however, RRnet = 2 or non-gamblers are recruited or both , the prevalence rises faster when the “exposure fraction” is higher, say 10% instead of 2.1%.

Even in a really catastrophic scenario like H, a sample more than three times larger than that of Welte et al would be needed to have a good chance of declaring that the difference between 3.99% and 4.83% is statistically significant. That difference, if true , would mean almost two million casualties that could might not be counted by research, but would be by two million families.

The different scenarios are summarized below. Recall, these are only a very few of an enormous set.


Scenario % pop who gamble Relative risk Are non-gamblers population prevalence
on internet of i-gambling enticed to i-gamble? of problem gambling
baseline 0.3 1 3.99
A 2.1 1 no 3.99
B 2.1 1 yes 4.04
C 2.1 2 no 4.11
D 2.1 2 yes 4.16
E 10 1 no 3.99
F 10 1 yes 4.26
G 10 2 no 4.56
H 10 2 yes 4.83

Those who advocate “regulation” (formerly known as “legalization”) of internet gambling are willing to let our country experiment with a technology whose effects on the population are not only unknown now, but cannot ever be known by research unless they are beyond catastrophic. e.g. doubling the prevalence of problem gambling nationally. Advocates of “regulation” like the representative of the Poker Players Alliance use the bad logic “absence of proof is proof of absence.”

It is shameful that policy-makers known for financial savvy pay no attention at all to elementary principles of public health. No rocket science is required, just a little caring.

Stephen Q. Shafer MD MPH 917 453 7371               8 Mynderse St.   Saugerties NY 12477

sqs1@columbia.edu or shpcount@earthlink.net

Permission is hereby given to reproduce the above in whole or part as long as the permalink is cited.