Q&A on Test 2: (Friday Nov. 14)
This page may be changing frequently on Thursday Nov. 13. If you are viewing this from your home computer using a graphical browser like Netscape or Internet Explorer, reload the page (click on reload) each time you look at it, or you might see a previously cached version rather than the latest one.
> You had mentioned that we have more than two enter > statements in the test. We had only done an example with two enter > statements and this is how it went; the first enter had the variables we > were controlling for, the second had the ones we were testing for with all > the other variables controlled. What would a third or fourth enter > statement test? Are we testing these variables controlling for all > variables mentioned in both enter statements or just controlling for > variables mentioned in the first enter statement. > > REGRESSION VARIABLES = STAY AGE NBEDS CENSUS NURSES INFRISK > /DEPENDENT = INFRISK > /ENTER = STAY AGE NBEDS > /ENTER = CENSUS > /ENTER = NURSES > > Does it mean: Testing for NURSES controlling for STAY AGE NBEDS CENSUS > YES! IN EACH F TEST FOR R-SQUARED CHANGE, YOU ARE CONTROLLING FOR THE VARIABLES MENTIONED IN ALL PREVIOUS ENTER STATEMENTS. ------------------------------------------------------------------- > When we are testing the significance of variables NOT included in the > model,(say 2 of them) the t test is done completely ignoring the 2nd > variable right? RIGHT. > But does this mean thal all the other variables that are > already in the model is controlled for or they are all also being ignored > when we are testing these 2 variables? > YOU ARE CONTROLLING FOR THE VARIABLES ALREADY IN THE MODEL. VARIABLES NOT IN THE MODEL ARE BEING TESTED ONE AT A TIME, CONTROLLING FOR THE VARIABLES ALREADY IN THE MODEL -- BUT NOT FOR THE OTHER VARIABLES NOT IN THE MODEL. THE OTHER VARIABLES "NOT IN THE MODEL" ARE BEING IGNORED. ------------------------------------------------------------------ > Planned comp. & Sceffe: even though the critical value at 0.05 for the > pooled t score (ie. sceffe value) is not given, can we conclude anything > from our t-value and t-prob for a planned comparison? > YOU WILL SEE NO SCHEFFE MATERIAL ON TEST 2, AND ONLY PAIRWISE SCHEFFE TESTS ON THE FINAL, AT MOST. BUT TO ANSWER YOUR QUESTION, ALL YOU CAN SAY IS THAT THE SCHEFFE CRITICAL VALUES ARE LARGER (MORE DEMANDING) THAN THE USUAL CRITICAL VALUES, UNLESS THE ONEWAY ANOVA IS COMPARING JUST 2 GROUPS, IN WHICH CASE THEY ARE THE SAME. SO I GUESS ALL YOU CAN SAY IN GENERAL IS THAT IF A PLANNED COMPARISON IS NOT SIGNIFICANT, NEITHER IS THE CORRESPONDING SCHEFFE TEST. ------------------------------------------------------------------ > 1, when do we have to use t-test and reduced / full model??? A T-TEST FOR A REGRESSION COEFFICIENT IS THE SAME (GIVES THE SAME P-VALUE AND IS INTERPRETED THE SAME WAY) AS IF YOU ENTERED JUST ONE VARIABLE IN A BLOCK AND LOOKED AT F FOR R-SQUARED CHANGE. ALL THE OTHER VARIABLES IN THE MODEL ARE BEING CONTROLLED (HELD CONSTANT). > like, if we say something is held constant or controlling for other > IVs. cause for me, the meaning of held constant and controlling are the > same thing. FOR ME, TOO, THEY ARE THE SAME. > By the way, can you explain the different of: "take in > account" , "ignoring" ... etc. TAKE INTO ACCOUNT = CONTROL = HOLD CONSTANT. THIS REFERS TO VARIABLES IN THE REDUCED MODEL. VARIABLES THAT ARE IGNORED ARE NOT IN THE FULL MODEL AND NOT IN THE REDUCED MODEL EITHER. > 2, Can you make up an example to show me what type of the question is > about t-test in regression and what type of the question is about the > reduced/full model??? Thanks. REGRESSION MODEL A: / ENTER CENSUS NURSES NBEDS / ENTER STAY. REGRESSION MODEL B: / ENTER CENSUS NURSES NBEDS STAY. IF YOU LOOK AT F FOR R-SQUARED CHANGE IN MODEL A, IT TELLS YOU EXACTLY THE SAME THING AS THE T-TEST FOR STAY IN MODEL B. IN FACT, F = T-SQUARED. IN BOTH CASES, YOU ARE TESTING FOR STAY, CONTROLLING FOR CENSUS, NURSES AND NBEDS. > > 3, If we held constant for the mean level, Do we always use the z-score > instead of the regression model?? Actually, how do we decide when do we > use z-score and when do we use regression model??? USING THE Z-SCORES GIVES YOU EXACTLY THE SAME CONCLUSIONS AS USING THE ORIGINAL INDEPENDENT VARIABLES. WE USE Z-SCORES BECAUSE HOLDING X CONSTANT AT THE MEAN LEVEL IS THE SAME AS HOLDING Z CONSTANT AT ZERO, AND IT'S EASIER TO PLUG ZEROS INTO A REGRESSION EQUATION THAN TO PLUG IN SAMPLE MEANS OF THE INDEPENDENT VARIABLES. BUT THE RESULTS ARE 100% THE SAME. > > 4, in the lecture notes Pg, 40. What do you mean the paragraph - "give > corrected mean......at its model level"? What is the corrected mean Y?? > IT MEANS HOLD MEDICAL SCHOOL AFFILIATION CONSTANT AT ITS MODE (THE MOST FREQUENT VALUE), WHICH IS NO. THE "CORRECTED MEAN" IS Y-HAT. > > P.S, when will you stop to check our e-mail??? > > THURSDAY NIGHT AROUND 9PM (THAT IS, AS SOON AS I HAVE FINISHED ANSWERING THIS BUNCH OF MESSAGES) -------------------------------------- > may I know if you said held costant of other variables, Do you mean we > set X1 x2.....etc to be a constant value (eg, 1, 2, 3....etc) or set these > var. equal to 0. what do you mean by ignoring? Under what condition we > have to use a Bo and under what condition we use B*? by the way, how can > we calculate the new B*? ZERO IS ONE POSSIBLE VALUE AT WHICH YOU MIGHT HOLD A VARIABLE CONSTANT. IN TERMS OF THE SLOPES, IT DOES NOT MATTER WHERE YOU HOLD THE VARS CONSTANT -- THEY ARE THE SAME EVERYWHERE IF THERE ARE NO INTERACTIONS. IF YOU HOLD THE VARIABLES CONSTANT AT THEIR MEAN LEVELS, IT IS THE SAME AS HOLDING THEIR STANDARDIZED VERSIONS (Z-SCORES) CONSTANT AT ZERO. IN THIS CASE, THE INTERCEPT OF THE REGRESSION EQUATION USING THE Z-SCORES IS EXACTLY B*. IN FACT, ALL THIS BUSINESS WITH Z-SCORES IS JUST TO SHOW YOU HOW TO MAKE SPSS CALCULATE THE B* FOR YOU. -------------------------------------- > Why when we compare the same DV and IV but in two different method - > oneway and regression(controlling for the other IVs), the outcome that I > get is totally different?? One is significant but the other is not > significant??? Why this happen?? and what do we have to conclude about > that? WHEN YOU DO A ONEWAY YOU ARE NOT CONTROLLING FOR ANYTHING. THE CONTROL VARIABLES CAN MAKE A BIG DIFFERENCE. LIKE FOR EXAMPLE, IF YOU WANT TO PREDICT NUMBER OF CHURCHES IN A CITY FROM NUMBER OF CRIMES, YOU WILL FIND THEY ARE POSITIVELY CORRELATED. BUT IF YOU FIRST CONTROL FOR POPULATION OF THE CITY, I BET YOU WILL FIND NO RELATIONSHIP. THE INTERPRETATION IS, FOR CITIES OF THE SAME SIZE (THAT IS, HOLDING POPULATION SIZE CONSTANT) CRIME AND RELIGION ARE UNRELATED -- WELL ANYWAY THE QUANTITY OF CRIME AND THE QUANTITY OF RELIGION ARE UNRELATED.
Well, that's it. I'm going to bed. Good luck on the test.