0

I'm learning statistics with SPSS. I have a test case where I need to analyze using regression linear method, I'm not sure if I can explain clearly so please refer to the picture. My understanding is to define the input into multiple-response set and then analyze the set using the regression linear method. Is my understanding on the problem wrong? Can someone please explain it if so? If my understanding is right, then how can I use the set as variable in regression linear? I use multiple-response option under analyze to define the set, is this the right way to do this?

enter image description here

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
Dizz
  • 3
  • 1
  • The picture looks not like a regression model but like a structural equation model with observed variables X1, Y11, Z1, etc. and latent variables 'Kualitas Sistem' etc. – hplieninger Jan 31 '18 at 09:12
  • I'm not sure because i haven't learn about structural equation model yet(also this is the first time i've seen case like this), but my professor said to analyze this using regression model. Can i analyze the structural equation model using spss? – Dizz Jan 31 '18 at 09:21
  • What i need to analyze is the significance from X to Y1, and from X,Y1,Y2 to Z – Dizz Jan 31 '18 at 09:24
  • Have you directly observed X, Y1, Y2, and Z? If not, then the picture resembles a SEM. If yes, then it's a path analysis. But you cannot do this with regression because Y1 is both an independent and a dependent variable. Search online and spend some time reading up on SEM. – hplieninger Jan 31 '18 at 09:31
  • 1
    No, i haven't observed that variable directly, forgot to mention the variables that contain data is only the small square x1, y11, z1, etc so 12 column in total. SEM as in Structural equation model right? Thanks i'll search it up on that. – Dizz Jan 31 '18 at 09:38
  • I read up a bit on SEM, my model does looks like that but how can i analyze it using spss statistics? If i understand correctly i need to use tools like spss amos to draw the model and analyze the data. – Dizz Jan 31 '18 at 16:02

1 Answers1

0

The picture resembles a structural equation model (), with manifest/observed variables X1, Y11, Y21, Z1, etc. and latent/unobserved variables X, Y1, Y2, and Z.

Regression can not handle the measurement model, i.e., the latent variable X measured by X1, X2, X3. And regression can not handle the structural model, i.e., the relationships among the latent variables; for example, Y1 is both an independent and a dependent variable.

A SEM can be fit using the SPSS Addon AMOS, but there is also and OpenMx (both free), as well as or Lisrel (both commercial), and many more.

Note further that the picture shows formative measurement models, that is, the arrows go from the observed to the latent variables. This may be intended, but what is more often seen are reflective measurement models, where the arrows go from latent to observed. (See, e.g., What is the difference between formative and reflective measurement models?.)

Last, SEM can be a complex topic, especially for novices. Read a book (maybe Raykov & Marcoulides), take an online or offline workshop, and tell your advisor that this is complex and this is not regression.

Raykov, T., & Marcoulides, G. A. (2006). A first course in structural equation modeling (2nd ed.). Mahwah, NJ: Lawrence Erlbaum.

hplieninger
  • 802
  • 9
  • 25
  • Thanks i'll tell and ask him in my class tomorrow. I still need to analyze this though, so for now my approach on this problem is like this: i transform the initial/observed variables by summing it up to become the latent variables, i.e.: i transform up x1, x2, and x3 to become X value. And then i use regression to analyze the results based on arrows on the picture. What do you think of my approach? I take this approach because the values of the observed variable is same between 1-5 for all of them. Sorry if i use wrong terminology. – Dizz Feb 01 '18 at 08:26
  • As I said, regression can not handle the situation that Y1 is a predictor for Z, but a dependent variable for X. With manifest variables, this is a path analysis (or mediation model), not regression. Don't split it up in two regression models, results will not be accurate. – hplieninger Feb 01 '18 at 09:01
  • Using the sum of the observed variables as a proxy for the latent variables will probably/hopefully do not much harm, but relationships among the latents will generally be closer to 0. – hplieninger Feb 01 '18 at 09:05
  • Ok, so this is positive that my professor gave wrong instructions. If i ignore the relation between X and Y1 will it make the result at least more relevant? – Dizz Feb 01 '18 at 09:11
  • It wont't hurt your estimates, but it won't give you the full picture either, because you then miss the indirect effect of X through Y1 on Z. – hplieninger Feb 01 '18 at 10:10
  • 1
    Thank you very much. I'll just ignore the relation for now because i need to at least submit something by tomorrow. I will make sure to confirm with him about what you said so i won't get confused again in the future. At least now i have a bit knowledge on SEM, maybe i'll go back to this case after i have better understanding on this. I tried trial version of AMOS before but the model is un-indetifiable, maybe there is something i miss. – Dizz Feb 01 '18 at 10:39
  • A little update. I ask my professor and he said that i do need to split up into two regression model to solve the case. When i ask about SEM he said it won't be given in this course, and i only need to do regression for this. As for why he use the model in the picture, i don't know, maybe he only want to test our basic understanding but that sure is confusing. – Dizz Feb 02 '18 at 13:14