 |
 | Home | Contact Us | FAQ | Search & Site Map | Link to Us |
 | Sign In | Join | Other 45 Sites in Network |
Math Forum / Math Software / SPSS / July 2009Help with multiple regression
Hello folks I am a physician who works in Mexico. I would like to predict the weight of a fetus before birth through ultrasound measurements. There are many studies which have published an equation or a formulae in order to estimate fetal weight, and the equation has been obtained from independent variables (parameters of ultrasound). Unfortunately, none of these studies has been done in Mexican population. I have collected the birth weight (dependent variable) of almost 500 newborns (NB). I hav also collected 13 ultrasound measurements (independent variables) per fetus in the 48 hours prior to birth (prenatal stage). My goal is to find an equation or formula to predict the weight of the baby using ultrasound variables (independent variables). I have read about this and I think I have to run a linear regression in which the dependent variable would be the birth weight, and ultrasound variables would be included as independent variables. According to what I have read, I have to choose a "selection of variables backwards method" by which I will obtain a linear model. The problem is that I have no experience on how to perform this, and after running the regression, at the results window I get a series of data such us tables (descriptive statistics, correlation, included/ deleted variables, a summary model, ANOVA, analysis of colinearity, excluded variables), and Graphics. What is the right way to run the multiple regression? How can I get the model from these data? Which data must be included in the equation? Thanks in advance for your help. PS: Sorry for my english > Hello folks > I am a physician who works in Mexico. I would like to predict the [quoted text clipped - 22 lines] > help. > PS: Sorry for my english It's hard to get fully what you intend but I am guessing you want to predict the birthweight based on a series of measurements with ultrasound. Could you be specific about what the measurements entail? Are they similar? Are there ways to combine some of them? You could use linear regression but you have a few issues to consider. One the residuals must be relatively normally distributed. Your continuous predictors should be centered and linear with the dep var (roughly). If your measures are highly related to each other then you will have multicollinearity. It sounds like this would be a problem here. You may have to find a way to combine some of them or eliminate some prior to running your regression I think the folks on this board could recommend some texts to consider. Tabachnick and Fidell is a reasonable and readable multivariate book with some complete examples of linear regresion. You could also seek out a consultant. Stepwise regression is not recommended for a few reasons, one being that it puts all of the decisions in your computer's hands. You should think carefully ahead of time, in clinical terms, about what the measurements represent, how they are related and when you should enter them into your equations. Best, Marc > Hello folks > I am a physician who works in Mexico. I would like to predict the [quoted text clipped - 3 lines] > from independent variables (parameters of ultrasound). Unfortunately, > none of these studies has been done in Mexican population. --- snip --- For those who are interested, there have been some replies to the same post in the Medstats Google group (http://groups.google.com/group/ MedStats). -- Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." You might want to try getting the predicted birth weights with your method and correlating them to the predicted birth weights from the other sources. You might also want to try using the same constructs to predict birth weight that previous studied did and compare and contrast the resulting equations. Art Kendall >> Hello folks >> I am a physician who works in Mexico. I would like to predict the [quoted text clipped - 15 lines] > http://sites.google.com/a/lakeheadu.ca/bweaver/ > "When all else fails, RTFM." >> Hello folks >> I am a physician who works in Mexico. I would like to predict the [quoted text clipped - 5 lines] > >--- snip --- Stepwise regression (in the snipped portion) has its special uses, but it is in low enough repute these days that some journals have guidelines stating that it generally should be avoided. >For those who are interested, there have been some replies to the same >post in the Medstats Google group (http://groups.google.com/group/ >MedStats). That URL gave me an error. I think this one will find the message, and then you can expand the thread. http://groups.google.com/group/MedStats/msg/683c0b0026811df9?hl=en As they suggest, too, I think you should start with validating what has been previously published. -- And then look to see if you can say anything about deviations from those prediction.  SignatureRich Ulrich To make sense out of the output (the results of your analysis), you should first see which variables are significant at predicting the birth weight (in the coefficients table). If you want to create a formula, you would technically multiply (for each baby) the bêta by the appropriate data point (say your bêta for ultrasound at -2 hours is .519, you would multiply the measurement at -2 hours by .519) and then add them together. That should (I think) give you the predicted weight of the baby. Hope this helps a little! |
Reply to this Message |
 Sign In
 Join
 My Latest Posts My Monitored Threads My Blog My Photo Gallery My Profile My Homepage Start New Thread Enable EMail Alerts Rate this Thread
|
©2010 Advenet LLC Privacy Policy - Terms of Use
This website includes both content owned or controlled by Advenet as well as content owned or controlled by third parties.