## Reports on bugs, errors and changes. |

December 11, 1997. SISA has started!

February 17, 1998, Improved the clear function for procedures. Added some comments to the top to make it easier accessible for Webcrawler users.

February 19, 1998, Changed the lay-out of texts. Input is now at the top instead of at the bottom.

March 3, 1998, Sorted some problems with the counter on the index page. Seems to be working fine now.

March 5, 1998, Fixed confusion in Negative Binomial vs1 both in the text and the actual program. Rediscovered that it doesn't work if mean and variance have the same value.

March 13, 1998, Improved error trapping in Ordinal (in Kolmogorov-Smirnov).

March 31, 1998, Both the htm and the pascal version of hypergeometric programs have been improved. Input and output are the same but the power has been increased.

May, 1998, The procedure random has been added to SISA. In the coming weeks we will make the following improvements: 1) Make the program downward compatible with Netscape 2.0; 2) Add some more methods such as an 8 bit generator and some non linear congruential generators; 3) Improve the explanation page.

Added L'Ecuyer and PaMi. Still having serious problems with Netscape 2.0 and 2.1. Basically, if you change the options box the thing will crash. So if you have these older versions of Netscape up-grade or only use MINSTD, **and don't touch the options box**. Another problem with Netscape 2.1 is that it doesn't underline properly. So mean+or-1.96 reads as mean+1.96 only. Thus, beware if you read the help text.

June, 1998, The provider installed a new server this month giving considerable disruption. '404's, weird things happening while downloading, and other problems, such as no e-mail and no FTP, were the result. Apologies for any inconvenience caused. The good thing is that the new server should be much faster and we are also going to get more web space for the same price.

June, 1998, Changes have been made to the 'Sample size' procedure. The input procedure has been made more sophisticated and the descriptions improved, we hope. 'Continuity correction' has been added. Please do not think you know what continuity correction means in this case, probably you don't, so please consult the manual. Also, a reference pertaining to the formulaes used has been added to the literature list (Sahai & Kurshid

June, 1998, The Bonferroni adjustment has been added.

Pernerger has been added to the literature listJune, 1998, Some changes have been made to help users of the more complicated procedures

June, 1998, The provider still has considerable problems with the new server

June, 1998, In the procedure 'Sample size' the allocation ratio procedure for proportions for the usual analysis has been changed. The previous procedure produced a conservative estimate suggesting slightly larger sample sizes than you actually needed. Other calculations in this procedure are not affected.

July, 1998. The 'population' procedure is added to the 'sample size' procedure. This is part of an ongoing upgrading and expansion of the sample size procedure. The t-test procedure will soon be upgraded to mirror the changes. It seems that the provider solved the server problems so everything should be fine on that front (although I think they are still a bit slow, particularly with CGI procedures).

July, 1998. The 'population' procedure is added to the 't-test' procedure. This is aimed to make the t-test procedure and the sample size procedure better mirror each other. Input of t-test procedure has been expanded and improved. Error trapping for both t-test and sample size has been slightly improved.

July, 1998. The procedure 'power' is added to the website. Machin et.al., on which most of this page, but not all, is based, is added to the reference page. Please note, there are errors in Formulae 4.1 and 4.2; SISA believes there is also an error in 3.5.

August, 1998. Some changes made to the NNT in the procedure t-test. Particularly introduced the two different ways to calculate the confidence interval around the NNT.

September, 1998. An error was discovered in the MsDOS program multinomial. If you have an old version of multinomial you must not use it but replace it. In the older versions the calculation for the point probability of the multinomial is not correct, all other output is O.K. (Many thanks to Lee Ping Yin from the People's Republic of China for pointing out this error). While fixing the problem we added a checksum procedure to the multinomial. The checksum procedure gives an indication of rounding errors and echoes the divergence from one, which should be zero. Normally one would expect the cumulative probability to be well within 'divergence value' from the value which is given for the cumulative probability.

October, 1998. Improvements have been made in calculating the confidence intervals for the risk-ratio and the odds-ratio in the procedures t-test and two by two table. A check box has been introduced to enable you to choose the confidence interval yourself. Estimates and procedures to analyze the difference between rates have been added to Poisson and Binomial. This will be slightly more extended in the future. SISA has now moved from Britain to the Netherlands

November, 1998. As is obviously clear there are problems with the checkbooks procedures for Netscape users. These users will have to manage with 95% confidence intervals only for the time being. Also, the procedure to truncate large numbers is a real pain for Netscape. We will look into that problem soon.

November, 1998. We did look into the problem and it seems that the problems only appears in older versions of Netscape. Netscape fixed the bugs and from version 4.0 onwards everything works fine. Therefore, Netscape users, upgrade.

November, 1998. The procedure Hypergeometric has been updated to allow a normal approximation with an error distribution which considers sampling without replication in a small population. The new procedure "*Simple*" has been added to do some really simple calculations which might come in handy in statistical analysis. If required simple will be extended to consider other simple calculations. Suggestions?

December, 1998. The index page has been changed to give some links to sites which might be interesting for SISA users. In the coming months we will start some new SISA services using this page. Those users who want to have nothing of SISA's new and slick appearance, please link to the instr.htm page which has not been changed.

June, 1999. Easter this year a page was produced with a listing of conferences. Unfortunately that did not go very well, so, the page will be downgraded while we contemplate what to do next.

June, 1999. There has been a consistent trickle of e-mails asking about how to calculate a sample size for a pairwise test. Well, it has been added to the procedure pairwise.

August, 1999. A bug has been discovered in the procedure two by two. The p-values of the Chi-squares and the correlation were not reliable. This bug is __only__ relevant for users of Netscape Navigator (and as far as I can see now only for the versions 4.0+). This bug did __not__ affect the functioning of Microsoft Explorer. The bug is now fixed and both Explorer and Navigator should function fine. Thanks to * Michael A Pannell* for drawing my attention to this problem.

August, 1999. The procedures for doing calculating the sample size for pairwise tests have been improved. Also the text has been improved. The text for the McNemar in the procedure pairwise has also been rewritten. Lets hope that everything is satisfactory now.

August, 1999. The power calculation for doing pairwise tests has been added to the procedure power, text has been added to the helpfile

August, 1999. Bugs were discovered in both the sample size and the power procedure for doing equality analysis (only!). These made that the procedures did not work satisfactory and unreliable results were produced, of the ridiculous kind so you should have recognized them. The bugs have been removed and extensive validation has been done. The equality analysis procedures now work fine.

August, 1999. The T-test procedure and the Sample size procedure have been removed in the DOS/Pascal download version. Basically these procedures have not been maintained lately and the web versions are now much more extensive using better estimation procedures. Should there be any burning needs for improved versions of these procedures let me know and I will see if I can provide upgraded versions of these procedures.

August, 1999. A slight change has been made in the way the significance of the correlation coefficient is calculated in the procedure *significance*. Instead of using a double sided Chi-square we now use a single sided t-test. Following Cohen. It is the more usual approach, so why not. There is very little difference between using the t-test as opposed to the Chi-square, as expected.

September, 1999. The procedure correlation has been added to the collection. Please give your comments and suggestions for improvement.

November 1999. Maybe someone has an answer to the question why SISA is no longer indexed in Webcrawler and Excite and only irregularly in AltaVista. I tried everything but no success. Remember people, Infoseek/Go and Google, yes try Google, are good and particularly reliable alternatives. Highly recommended!

November 1999. A new procedure reverse significance is introduced. I imagine that you can use it to teach the students to find the value of 'z' which is related to 0.025.

December 1999. For the procedure significance a new estimation procedure is used to estimate z very efficiently and separate from the t-test. The F-value procedure is exchanged for a better one. Small improvements have been made to the Chi-square and the t-tests. Everything should now have at least six-digit precision.

December 1999. To prepare for the SMR procedure I studied a few algorithms to estimate the Gamma and the Beta function. It was a depressing and time wasting experience. The results of my study have been put in the module Gamma & Beta, which is not very flexible but at least provides reliable results. In the future I will look at some more algorithms and I gradually hope to improve the procedure.

December 1999. Introduced the SMR-Exact procedure.

October 2000. Brent Meeker wrote the continuity correction and exact p-value procedures for the Wald-Wolfowitz Runs test.

December 2000. An error was found in an early version of the MS Windows Tables program Exact analysis. This error was fixed on December 1, 2000.