Hey,
For my research I have multiple independent and multiple dependent variables (and some covariates). One independent variable has four categories while the others are continuous. The dependent variables are all continuous. I think the covariates should be continuous but apart from continuous variables I also want to test for the effect of categorical variables like gender and employment status (how should I do this?) My guess was to try MANCOVA, so I did this with the independent variable and the 4 groups, and the dependent variables + covariates. It worked (except that I don't know if it is appropriate to put gender and employment status etc. under covariates). I then added the continuous independent variables to the test, but it didn't work (some warning that it was too much). - How can I deal with gender and employment status? - Should I use a different test for the analyses, or just for analysing the effect of the continuous independent variables on the dependent ones? Thank you!! -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
You can use a multiple regression approach dummy coding the categorical independent variables. Covariates are entered as blocks, as in a heirarchical multiple regression. There's a tutorial at:
Brian
From: SPSSX(r) Discussion <[hidden email]> on behalf of spssdummy <[hidden email]>
Sent: Monday, June 10, 2019 11:21:34 AM To: [hidden email] Subject: test with multiple independent and dependent variables Hey,
For my research I have multiple independent and multiple dependent variables (and some covariates). One independent variable has four categories while the others are continuous. The dependent variables are all continuous. I think the covariates should be continuous but apart from continuous variables I also want to test for the effect of categorical variables like gender and employment status (how should I do this?) My guess was to try MANCOVA, so I did this with the independent variable and the 4 groups, and the dependent variables + covariates. It worked (except that I don't know if it is appropriate to put gender and employment status etc. under covariates). I then added the continuous independent variables to the test, but it didn't work (some warning that it was too much). - How can I deal with gender and employment status? - Should I use a different test for the analyses, or just for analysing the effect of the continuous independent variables on the dependent ones? Thank you!! -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Thanks Brian!
Very useful tutorial. The only thing is that this is suitable for just one dependent variable at a time right? So should I do the same multiple regression analysis multiple times as I have multiple dependent variables? Second, my independent categorical variable has number 1, 2, 3 and 4 right now. Is this okay for the analyses? Thanks again -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Your 4-category independent variable is fine. It will require three dummy variables. You'll probably need to look up how to dummy code multiple-category variables if you don't know how at this time. Multiple regression is made for one dependent variable. There is an alternative called canonical correlation (notice the term correlation connoting that you cannot attribute independence/dependence). Canonical correlation finds the best set of variables in set A (e.g., your independent variables) to the best in set B (e.g., your dependent variables). So basically, you have the correlation between one composite variable and another composite variable. Canonical correlation is a real bear to interpret. I'd stick with one dependent at a time.
Brian
From: SPSSX(r) Discussion <[hidden email]> on behalf of spssdummy <[hidden email]>
Sent: Monday, June 10, 2019 2:41:59 PM To: [hidden email] Subject: Re: test with multiple independent and dependent variables Thanks Brian!
Very useful tutorial. The only thing is that this is suitable for just one dependent variable at a time right? So should I do the same multiple regression analysis multiple times as I have multiple dependent variables? Second, my independent categorical variable has number 1, 2, 3 and 4 right now. Is this okay for the analyses? Thanks again -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
For creating a set of dummy variables from a categorical variable, check out the SPSSINC CREATE DUMMIES extension command installable from the Extensions > Extension Hub menu. It will appear on the Transform menu once installed. If you do want to venture into canonical correlations, there are several ways to get them in Statistics, but the easiest is probably the STATS CANCORR extension command, also installable from the Extension Hub. On Mon, Jun 10, 2019 at 12:53 PM Dates, Brian <[hidden email]> wrote:
|
In reply to this post by bdates
Thanks! I will then dummy code the variable. However, how can I get any more
insight in the specific linear relationships? Can I get something like multiple comparisons with multiple regression? -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Administrator
|
In reply to this post by spssdummy
Do you mean that you are using GLM MULTIVARIATE? Via the GUI, this would be
Analyze > General Linear Model > Multivariate, and the first dialog that appears would look like this: https://statistics.laerd.com/spss-tutorials/img/one-way-MANOVA-3.png If this is what you mean, note that Fixed Factors are categorical explanatory variables, and Covariates are quantitative explanatory variables. For the Fixed Factors, SPSS takes care of the dummy-coding that Brian mentioned automatically. (The REGRESSION makes no distinction between fixed factors and covariates. It treats all explanatory variables as if they are quantitative. So if you wish to include categorical variables when using REGRESSION, you do need to generate your own dummy variables.) One other thing you might want to think about is whether you have one multivariate question or several univariate questions (i.e., one question per DV). See the classic article by Huberty & Morris (1989) for more info about that. https://pdfs.semanticscholar.org/ee6c/77c99c8e4530d0cccaedf85ed525fb22a02d.pdf HTH. spssdummy wrote > Hey, > For my research I have multiple independent and multiple dependent > variables > (and some covariates). One independent variable has four categories while > the others are continuous. The dependent variables are all continuous. I > think the covariates should be continuous but apart from continuous > variables I also want to test for the effect of categorical variables like > gender and employment status (how should I do this?) > > My guess was to try MANCOVA, so I did this with the independent variable > and > the 4 groups, and the dependent variables + covariates. It worked (except > that I don't know if it is appropriate to put gender and employment status > etc. under covariates). I then added the continuous independent variables > to > the test, but it didn't work (some warning that it was too much). > > - How can I deal with gender and employment status? > - Should I use a different test for the analyses, or just for analysing > the > effect of the continuous independent variables on the dependent ones? > > Thank you!! > > > > -- > Sent from: http://spssx-discussion.1045642.n5.nabble.com/ > > ===================== > To manage your subscription to SPSSX-L, send a message to > LISTSERV@.UGA > (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
I want to know the effect of season (4 groups) and weather conditions
(multiple continuous independent variables) on physical activity variables (continuous). For example, is there an effect of season on sedentary time, or is there an effect of season on standing time, is there an effect of sun hours on low intensity activity? I was first trying glm-->multivariate but it didn't work with the independent variables weather conditions, when I tried to add these to the independent variable season. Brian advised me to use multiple regression (analyze-->regression) so that's what I am trying now. Therefore I wanted to know if you can get something like multiple comparisons with multiple regression. Thank you -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Administrator
|
What does "it didn't work" mean? My guess would be that one or more
variables got kicked out of the model due to multicollinearity. E.g., there must be a strong association between sun hours and season. Look at the figure on this page showing sun hours by month, for example: https://www.livingin-canada.com/sunshine-hours-canada.html Also, judging by the example questions you give below ("is there an effect of season on sedentary time, or is there an effect of season on standing time, is there an effect of sun hours on low intensity activity?"), I'd say you have a bunch of univariate questions, not one multivariate question. HTH. spssdummy wrote > I want to know the effect of season (4 groups) and weather conditions > (multiple continuous independent variables) on physical activity variables > (continuous). For example, is there an effect of season on sedentary time, > or is there an effect of season on standing time, is there an effect of > sun > hours on low intensity activity? > > I was first trying glm-->multivariate but it didn't work with the > independent variables weather conditions, when I tried to add these to the > independent variable season. > Brian advised me to use multiple regression (analyze-->regression) so > that's > what I am trying now. Therefore I wanted to know if you can get something > like multiple comparisons with multiple regression. > > Thank you > > > > -- > Sent from: http://spssx-discussion.1045642.n5.nabble.com/ > > ===================== > To manage your subscription to SPSSX-L, send a message to > LISTSERV@.UGA > (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
In reply to this post by spssdummy
One reason that manova might "not work" is the lack of d.f.
for the analysis. For multiple regression, you have several d.f.
on the right hand side of the equation, only 1 d.f. on the left.
For any analysis at all, you need at least as many cases as there
are d.f., and it is better to have twice, five times, ten times, ...
depending on the size of the underlying R2 you expect in the model.
Remember - for simple MR, the R2 achieved by /artifact/, by
over-prediction, is (predictor d.f.)/cases. The actual prediction
has to be visible against that magnitude of "noise".
In my few exercises with manova, I've figured the same rule of
thumb for N seems okay, but this time, taking the "total d.f." as
(d.f.-left side) times (d.f.-right side). And that comes out too big
for small data sets.
With the problem that you eventually describe, it seems to me
that if you want a /test/, you want to create composite scores -
probably for both dep. and indep. vars. That reduces the test to
multiple regression, with fewer variables. If this "overall test" is
good, then you are justified in exploring its components.
If you are not choosing manova for its /test/ - Is there a proper
rationale for using it at all? Do you have this hypothesis, that you
are looking for a useful, not-obvious contrast among the dependent
variables?
Manova is notable for being hard to interpret. The complications
of figuring out "confounding", etc., among the independent vars
in multiple regression are magnified by simultaneously trying to
figure out what is happening among the dependent variables.
(Not really bad for 2; "it depends" for 3; probably bad for 4 or more
dep vars. But may be okay if you totally know what you are doing.)
--
Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of spssdummy <[hidden email]>
Sent: Monday, June 10, 2019 11:21 AM To: [hidden email] Subject: test with multiple independent and dependent variables Hey,
For my research I have multiple independent and multiple dependent variables (and some covariates). One independent variable has four categories while the others are continuous. The dependent variables are all continuous. I think the covariates should be continuous but apart from continuous variables I also want to test for the effect of categorical variables like gender and employment status (how should I do this?) My guess was to try MANCOVA, so I did this with the independent variable and the 4 groups, and the dependent variables + covariates. It worked (except that I don't know if it is appropriate to put gender and employment status etc. under covariates). I then added the continuous independent variables to the test, but it didn't work (some warning that it was too much). - How can I deal with gender and employment status? - Should I use a different test for the analyses, or just for analysing the effect of the continuous independent variables on the dependent ones? Thank you!! -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In reply to this post by Bruce Weaver
Thanks! There are indeed multiple questions but I do try to see it as one
question per dependent variable so 5 in total in my case. If I have the dependent variable sedentary time, i want to examine which independent variables affect it so I add all the independent variables (season, weather conditions) in the multiple regression model and do this again for each dependent variable if I'm correct Then I do like to know in what way exactly for example season affects the dependent variable. I might find an effect but with dummy coding all groups are only compared to the reference while i want to compare all groups to each other, which I don't know how to achieve with multiple regression. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
If I understand the question/issue correctly, you just simply run a series
of regressions where you alter which is the excluded reference category. The only meaningful thing that will change, however, is the p values since the rest is all algebraically equivalent. There's an option to use contrast coding, but I think that introduces a complexity that I don't think you need. Jeff -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of spssdummy Sent: Tuesday, June 11, 2019 7:49 PM To: [hidden email] Subject: Re: test with multiple independent and dependent variables Thanks! There are indeed multiple questions but I do try to see it as one question per dependent variable so 5 in total in my case. If I have the dependent variable sedentary time, i want to examine which independent variables affect it so I add all the independent variables (season, weather conditions) in the multiple regression model and do this again for each dependent variable if I'm correct Then I do like to know in what way exactly for example season affects the dependent variable. I might find an effect but with dummy coding all groups are only compared to the reference while i want to compare all groups to each other, which I don't know how to achieve with multiple regression. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Administrator
|
In reply to this post by spssdummy
Use GLM/UNIANOVA rather than REGRESSION and include categorical variables as
Fixed Factors. Then use EMMEANS with COMPARE keyword to get a table showing all pair-wise comparisons. You may also wish to use one of the ADJ() options to adjust for multiplicity. https://www.ibm.com/support/knowledgecenter/en/SSLVMB_25.0.0/statistics_reference_project_ddita/spss/base/syn_unianova_emmeans.html spssdummy wrote > Thanks! There are indeed multiple questions but I do try to see it as one > question per dependent variable so 5 in total in my case. If I have the > dependent variable sedentary time, i want to examine which independent > variables affect it so I add all the independent variables (season, > weather > conditions) in the multiple regression model and do this again for each > dependent variable if I'm correct > > Then I do like to know in what way exactly for example season affects the > dependent variable. I might find an effect but with dummy coding all > groups > are only compared to the reference while i want to compare all groups to > each other, which I don't know how to achieve with multiple regression. > > > > -- > Sent from: http://spssx-discussion.1045642.n5.nabble.com/ > > ===================== > To manage your subscription to SPSSX-L, send a message to > LISTSERV@.UGA > (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
In reply to this post by Jeff A
Thank you! I still wanted to ask how to interpret the results for the dummy
variables season. In my attached file summer is used as the reference and then there are the standardized beta coefficients but I am not sure if these are actually like a comparison to the other seasons (so the ones next to spring, autumn, winter). <http://spssx-discussion.1045642.n5.nabble.com/file/t341602/spss.png> -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In the sample, Autumn has the lowest predicted value on the DV followed by
Winter, Spring, and then Summer. You have little evidence that there is a non-zero difference in the DV across any of the seasons in the population from which the sample is drawn. ...many might say that none of these differences reaches the traditional .05 level of statistical significance. You don't really need to do further comparisons if you're simply interested in null-hypothesis-statistical-tests, since if there is no meaningful population difference between the three included seasons and the summer reference where all three included seasons are less than summer in the sample, then they can't be meaningfully different from one another - E.g. if the Autumn/Summer p value = .146 and Autumn has the largest difference from Summer, then all other contrasts/comparisons will have p>.146. Trying to interpret standardized coefficients here doesn't seem very useful or meaningful to me. -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of spssdummy Sent: Tuesday, June 11, 2019 11:12 PM To: [hidden email] Subject: Re: test with multiple independent and dependent variables Thank you! I still wanted to ask how to interpret the results for the dummy variables season. In my attached file summer is used as the reference and then there are the standardized beta coefficients but I am not sure if these are actually like a comparison to the other seasons (so the ones next to spring, autumn, winter). <http://spssx-discussion.1045642.n5.nabble.com/file/t341602/spss.png> -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Thanks! Indeed for this dependent variable no further analyses will be
needed. However, when looking at the descriptives, the values for the dependent variables seem to be different, with the highest value for winter, and then autumn, summer and spring are lower. So I don't know why the predicted results are so different. <http://spssx-discussion.1045642.n5.nabble.com/file/t341602/spss2.png> -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Administrator
|
In reply to this post by Jeff A
There is no need to estimate the same model multiple times to get all of the
desired pair-wise contrasts. One can use GLM/UNIANOVA (with EMMEANS) rather than REGRESSION, as suggested here: http://spssx-discussion.1045642.n5.nabble.com/test-with-multiple-independent-and-dependent-variables-tp5738046p5738060.html If a table of coefficients is desired, one can include PARAMETER on the PRINT sub-command. Jeff-2 wrote > If I understand the question/issue correctly, you just simply run a series > of regressions where you alter which is the excluded reference category. > The > only meaningful thing that will change, however, is the p values since the > rest is all algebraically equivalent. There's an option to use contrast > coding, but I think that introduces a complexity that I don't think you > need. > > Jeff ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
I tried that but SPSS couldn't handle it for some reason. 5 dependent
variables, 1 categorical independent (4 groups), 7 continuous independent variables, and multiple categorical and continuous covariates were entered. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In reply to this post by spssdummy
...because you have control variables in the multiple regression that are
correlated with the seasons. For one, you have max temperature that's going to be highly correlated with the season essentially by definition to the point that the entire model is likely meaningless. If all control variables have zero correlation with the seasons, then the descriptives will be equivalent to the regression. Any correlation between the controls and the season dummies will alter the season's coefficients (relative to the descriptive equivalents), and too much correlation will likely make the regression essentially meaningless if you're attempting to interpret the model rather than simply make a prediction for the sake of pure prediction and nothing else. Look up info on the collinearity statistics you've printed out and look up very fundamental information on multiple regression and control variables in an intro stat book. What you're doing (for simplistic example) is essentially like asking the software to make a prediction about what is expected to occur during the summer at times when the seven day average max temperature is zero or in the winter when it is 38 C and expecting the results to be meaningful or able to be interpreted in a meaningful way. It would likely be more helpful to read up on basic statistics and ask about those principles in the appropriate place rather than ask about running moderately-advanced models on this list when what you're running sounds meaningless to me. To me, it makes little sense to know the steps in running a model in statistical software if you don't know what it means. -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of spssdummy Sent: Wednesday, June 12, 2019 12:04 AM To: [hidden email] Subject: Re: test with multiple independent and dependent variables Thanks! Indeed for this dependent variable no further analyses will be needed. However, when looking at the descriptives, the values for the dependent variables seem to be different, with the highest value for winter, and then autumn, summer and spring are lower. So I don't know why the predicted results are so different. <http://spssx-discussion.1045642.n5.nabble.com/file/t341602/spss2.png> -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In reply to this post by Bruce Weaver
Thanks. I wasn't following the thread closely but jumped in the middle to
see what all of the discussion was about ! ...will have to look into GLM/Unianova sometime - not familiar with that procedure. Given the nature of the original poster's questions, however, I'm unsure whether it makes sense to suggest that he tries more models that he can't interpret and are probably meaningless ! -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Bruce Weaver Sent: Wednesday, June 12, 2019 6:18 AM To: [hidden email] Subject: Re: test with multiple independent and dependent variables There is no need to estimate the same model multiple times to get all of the desired pair-wise contrasts. One can use GLM/UNIANOVA (with EMMEANS) rather than REGRESSION, as suggested here: http://spssx-discussion.1045642.n5.nabble.com/test-with-multiple-independent -and-dependent-variables-tp5738046p5738060.html If a table of coefficients is desired, one can include PARAMETER on the PRINT sub-command. Jeff-2 wrote > If I understand the question/issue correctly, you just simply run a > series of regressions where you alter which is the excluded reference category. > The > only meaningful thing that will change, however, is the p values since > the rest is all algebraically equivalent. There's an option to use > contrast coding, but I think that introduces a complexity that I don't > think you need. > > Jeff ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Free forum by Nabble | Edit this page |