Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
Drexel University or The Math Forum.



[apcalculus] Re: logistic regression on the TI86 (and TI89)
Posted:
Mar 27, 2001 7:23 PM


Drew and others who have been responding to this topic 
The problem is that logistic regression has what the fractals and chaos people call "sensitive dependence on intial conditions." Try entering data points that are well to the left of the point of inflection and doing logistic regression. Then make a small change in the last data point. The location of the upper horizontal asymptote can change by a factor of 2 to 10. What appears to be happening with the TI86 that allows a vertical shift as well as the other three constants seems to be related to this sensitive dependence on initial conditions.
The moral of the story is, don't place too much reliance on predictions of such things as the ultimate population of the United States based on logistic regression. The usual cautions about extrapolation apply here, too.
Regards, Paul
 From: Drew Isola <disola@accn.org> To: "APCalculus" <apcalculus@list.collegeboard.org> Subject: [apcalculus] Re: logistic regression on the TI86 (and TI89) Date: Mon, Mar 26, 2001, 2:32 PM
I realize that the 86 regression equation allows for a vertical shift but shouldn't the equation give values that are at least close to the 83 equation since they were generated from the same data?
Drew
Roger Keil wrote: I agree with the comment about different solutions but not "errors". If you look at the equations the TI 86 allows for a vertical shift that the 83 does not take into consideration. It actually allows a more complete solution then the 83. Also I believe the set of data you are using trys to take into account the ln of "zero" which would cause the problem.
Roger Keil North Hunterdon High School Annandale NJ
Drew Isola wrote: We got the same errors also, but we are only using 86's and 83's. This happened when we tried to do #23 &24 on page 348 of FDWK, the same section you cited.
Drew
Greg Kelly wrote: We are using the TI89, and the TI89 never gets the same regression equations as the book. When finding the logistic regression equation for a set of data, I sometimes get a message that says "Error: Singular Matrix". At other times everything works fine, and I can't find a pattern as to why it works sometimes and not others. Can anyone tell me what is going on?
The data that I am having trouble with comes from the example problem on page 345 of "Calculus: Graphical,Numeric, Algebraic" by Finney, Demana, Waits & Kennedy. The TI83 does the problem successfully, but the TI86 and TI89will not. Interestingly enough, the TI89 will solve the problem if I leave off the first data point.
This data produces the error: L1:{0,10,20,30,40,50,60,70,80,90,100,110} L2:{50.2,63,76,92,105.7,122.8,131.7,151.3,179.3,203.3,226.5,248.7}
Thanks, Greg Kelly, Hanford High School, Richland, Washington
 You are currently subscribed to apcalculus as: disola@accn.org To unsubscribe send a blank email to leaveapcalculus5913831W@list.collegeboard.org
To access the list archives or change your settings, visit http://list.collegeboard.org/cgibin/lyris.pl?site=collegeboard&enter=apcal culus  You are currently subscribed to apcalculus as: keilrj@ptdprolog.net To unsubscribe send a blank email to leaveapcalculus5913831W@list.collegeboard.org
To access the list archives or change your settings, visit http://list.collegeboard.org/cgibin/lyris.pl?site=collegeboard&enter=apcal culus  You are currently subscribed to apcalculus as: disola@accn.org To unsubscribe send a blank email to leaveapcalculus5913831W@list.collegeboard.org
To access the list archives or change your settings, visit http://list.collegeboard.org/cgibin/lyris.pl?site=collegeboard&enter=apcal culus  You are currently subscribed to apcalculus as: foerster@idworld.net To unsubscribe send a blank email to leaveapcalculus5913831W@list.collegeboard.org
To access the list archives or change your settings, visit http://list.collegeboard.org/cgibin/lyris.pl?site=collegeboard&enter=apcal culus
 You are currently subscribed to apcalculus as: archive@mathforum.com To unsubscribe send a blank email to leaveapcalculus5913831W@list.collegeboard.org
To access the list archives or change your settings, visit http://list.collegeboard.org/cgibin/lyris.pl?site=collegeboard&enter=apcalculus



