SOC 206                                                                                                                 Spring 07

Second Assignment

Part One:

In Part One you will estimate a path model of your choice without latent variables.

  1. The preparation of the model

 

  1. Pick a dependent variable, preferably one that is either measured at the interval/ratio scale, or one that is ordinal and has at least 5 categories.
  2. Select at least three independent variables. (The selection can be identical to the one that you made for assignment 1). Draw the path model. Don't forget the errors in the endogenous variables. Draw in all paths, but make sure that your model is recursive. Also, only exogenous variables can be connected with double headed arrows. Make sure that all endogenous variables are interval/ratio or ordinal with 5 categories. Only exogenous variables can be dichotomies.
  3. Write in the direction (+ or -) of the direct effect over each arrow.

 

  1. Preparation of the data

You will have to prepare the data in SPSS. This will involve running the regression first in SPSS to be able to check your AMOS run, then running the bivariate correlation to see the total associations, and finally, creating a smaller file that has only the variables you want to work with and only the cases that have no missing value on any of these variables. I provide an example that you can adapt to your variables.

 

*Getting the data file.

get file 'filename.sav'.

*Recodes if necessary, make sure to recode all missing as sysmis.

recode reliten (lo thru 0 = sysmis) (4 thru hi=sysmis).

recode absingle (2=1) (1=2).

*Selecting cases with no missing values on any of the variables included in the model THIS IS NECESSARY ONLY BECAUE WE ARE USING AMOS LATER.

*The following transformations will each create a dummy which takes 1 when the case is system missing or missing on the variable and 0 otherwise.

comp xeduc=sysmis(educ).

comp xxeduc=missing(educ).

comp xpremar=sysmis(premarsx).

comp xxpremar=missing(premarsx).

comp xreliten=sysmis( reliten).

comp xxreliten=missing(reliten).

comp xabsing=sysmis(absingle).

comp xxabsing=missing(absingle).

*To check these new dummies.

freq educ xeduc xxeduc premarsx xpremar xxpremar reliten xreliten xxreliten absingle xabsing xxabsing.

*Selection of the non-missing cases.

select if (xeduc ne 1 and xxeduc ne 1 and xpremar ne 1 and xxpremar ne 1 and  xreliten ne 1 and xxreliten ne 1 and xxabsing ne 1 and xabsing ne 1).

*Check for the selection.

freq educ premarsx reliten absingle.

*Regression to check against later.

REGRESSION

/MISSING LISTWISE

/STATISTICS COEFF OUTS R ANOVA

/CRITERIA=PIN(.05) POUT(.10)

/NOORIGIN

/DEPENDENT absingle

/METHOD=ENTER reliten premarsx educ.

*Correlation matrix to see the total associations.

CORRELATIONS

/VARIABLES=absingle premarsx educ reliten

/PRINT=TWOTAIL SIG

/MISSING=PAIRWISE .

*Save the file with only the variables you need.

save outfile 'newfile.sav' /keep=educ premarsx reliten absingle.

 

  1. Running AMOS

To run AMOS you have to click on Amos Graphic in the the Program Menu. Getting the output will involve three steps. Step A. you will have to specify your data file. Step B. You will specify the model by drawing the path diagram. Step C. you will have to estimate the model. .

A. Specifying the data file

Click on File on the main toolbar. From the pull-down menu select �Data Files�. You will see a window. Click on �File name� and choose your data file (newfile.sav). You will see in the internal window Group number 1, the name of the file you chose  and the number or cases usable/number of cases overall. Click OK.

B. Drawing the path diagram

  1. Draw the observed variables.

On the left, you will find a window with various tool icons. Click on the  rectangular icon to create the rectangles for each observed variable.  Use the photocopy icon to make  rectangles of the same size. If you need to move  a rectangle, change its size or delete it, right click on the mouse and choose the right command from the pull-down menu (you can also click on the right icon in the tool-icons window). .

  1. Draw the unobserved variables

Remember that each endogenous variable has an error term. The error term is not observed directly (it does not exist outside the particular model), but calculated and is thus an "unobserved" variable. Click on the oval shape icon  in the tool-icons window   Create the ovals for each unobserved (error) variable. Use the photocopy icon to make ovals  of the same size.  If you need to move an oval, change its size or delete it, right click on the mouse and choose the right command from the pull-down menu (you can also click on the right icon in the tool-icons window).  Another way of moving an item is by using the little truck from the tool window.
Shortcuts. In the tool window, there is an icon for the error term and also one for unobserved variables with observed indicators..

 

  1. Name the variables

Double click  on one of the rectangles or ovals . You will see a window pop up. Choose text and write  the variable name for that variable in the �Variable Name� window (the names must be identical to the ones you have in the new SPSS file!). Alternatively, to specify the names of the variables in your rectangulars, you can click on View/Set in the main toolbar. From the pull-down menu select �Variables in Dataset�. This will open a window with all the variables in the data set that you have specified in step A.  Drag each of these variable-names to its rectangular. . (The unobserved variables can be called error1, error2 etc., or anything else, that makes sense to you).

  1. Draw the causal paths.

Click on  the arrow shape icon in the tool-icons window . Click on the independent variable and while holding the mouse button down extend the line to the dependent variable and release button. Don't forget to draw a path from each (unobserved) error variable to its own dependent variable pointing in the right direction (from error to dep.var.!).

  1. If you have more than one exogenous variables, draw the unanalyzed paths (double headed arrows).

Click on the double-headed arrow shape icon in the tool-icons window Point at the exogenous variable lower on the chart and extend the curve to the exogenous variable above it.

  1. Unobserved variables need to be given a metric or scale to be estimable. We usually measure them by the same units the dependent variable is measured in. AMOS, however, needs to be told this. So we have to fix the metric (unstandardized) coefficient of the error terms to 1.

Double click on one of the path from the error to the dependent variable. You will see a window pop up. Choose parameters and write �1� in the �Regression weight� window. Close the window. In your path diagram you will see the number �1� assigned to the path that goes from the oval variable. 

 

NOW YOU ARE READY TO ESTIMATE YOUR MODEL.

 

C. Estimating the model.

 

 

 

i. Prepare the output

Click on �View/Set� on the main toolbar. From the pull-down menu select �Analysis Properties�. A window will pop-out. Choose �Output�. Specify all the choices on the right column. Specify the first five choices on the left. 

iiRun the estimation.

Click on Model-Fit. From the pull-down menu select Calculate Estimates. It will ask you for a filename with the .amw extension. This is where the content of your path diagram will be saved. The output will be written into a file with the same name you give here, except with an extension .amo.

iii Get the standardized coefficients on your diagram.

Amos finishes running the estimations when the icon with the point-up arrow in the left upper corner is highlighted in red Click on this icon to get the coefficients on your diagram. These are the unstandardized coefficient. To get the standardized coefficient click on �standardized estimates� in the small window at the left side of the path diagram. Printing your output

  1. You first print your diagram with the coefficients.

Make sure you have your model with the right coefficient on the top. Click on Edit on the main toolbar. From the pull-down menu select Fit to Page. Then click on File on the main toolbar. From the pull-down menu select Print. Then print the standardized output. You can also copy the diagram and past it to another document (e.g. Word) by selecting Edit/Copy to Clipboard.

  1. Then you print your text output.

Your text output will be in a file ending in .amo. To open the Output file click on View/Set. From the pull-down menu select �Text Output�.  Choose File from the top left corner and select �Print�.

Save your files before you exit. There will be four new files created by AMOS. *.amw will be the file that you need to call up if you want to redo the analysis. It is a command file with that includes your diagram. *.amo is your output file. *.ami or *.amj is the command file that you would have written had you not used point and click. *.amp is your technical output

 

Look at your output and compare it with the regression output from SPSS.

Then redo the analysis but now constrain all non-significant parameters (with C.R. less than 1.96) to 0 (the way you constrained the errors to have metric coefficients of 1). See 3. A. vi. Instead of setting the regression weight to 1, click on the desired path and set the regression weight to 0. If all your parameters are significant, choose the one with the lowest C.R. and constrain that to 0.

 

Answer the following questions:

  1. What did you learn from just looking at the direct effects?
  2. Is there a variable where the indirect effect is considerable?
  3. How much of the variation in your dependent variables are you explaining? (Pay special attention to the final dependent variable.)
  4. What happened to your model after you set some of the paths to zero? Did the other paths change as well? How good is the fit?
  5. Look at the two variables at the two ends of the deleted path. Go back to the first model with the path in. Write out all the different ways those two variables are related using the path coefficients from the first analysis. Then redo the same thing using the path coefficients from the second analysis, where the direct effect is set to 0. What is the total association in the first and in the second case?


HAND IN YOUR OUTPUT FROM SPSS AND AMOS with your answers..

 

Part Two

In Part Two you do a latent variable analysis.
For this you will need to download a data set from the US portion of the World Value Survey (CH11WVS.SAV. right click on this link and choose Save Link Target As). If you wish you may create your own data set and your own latent variable(s) for this part of the assignment.

The question we want to answer is how happiness is influenced by individualist ideology. Are people who subscribe to a more individualistic ideology happier? Both happiness and individualistic ideology are hard to capture and will be modeled as a latent variable. There is a third, control variable, gender.

happiness, will have three indicators. Each scored on a scale from 1-10 where 1= completely dissatisfied and 10=perfectly satisfied.

:

HAPHOME (V180) Overall, how satisfied are you with your home life?

HAPFIN (V132) How satisfied are you with the financial situation of your household?

HAPLIFE (V96) All things considered, how satisfied are you with your life as a whole?

 

individualism, will have four indicators. Each is scored on a ten point scale of agreement.

1              2              3              4              5              6              7              8              9              10

PRIVOWN (V251)

Government ownership of                                                  Private ownership of

business and industry should be increased                    business and industry should be increased

INDRESP (V252)

The state should take more                responsibility                       Individuals should take more responsibility

to ensure that everyone is provided for                           to ensure that everyone is provided for

COMPETE (V254)

Competition is harmful. It brings out the                         Competition is good it stimulates people to

worst in people                                                                     work hard and develop new ideas

HARDWRK (V255)

Hard work doesn't generally bring success                     In the long run, hard work usually brings a

-- it is more a matter of luck and connections                  better life

GENDER                1=Male and 2=Female

Create the proper model. A few hints:

Select the icon for the latent variable in the tools-icon window.  Drew oval shapes for the two latent variables or factors (Happiness and Individualism). To draw the indicators, click on  each of the ovals  as many times as many indicators it has.

Don't forget that since Individualism and Gender are causes and Happiness is effect, there is an error term in the factor Happiness, that you must draw as a latent variable with a path pointing from the error term to the factor with its parameter constrained to 1. The same goes for Individualism that is causally influenced by Gender.

Moreover, you will notice that one of the indicators will have a path from the latent variable that is fixed to 1. This is necessary, because this fixes the metric of this underlying, latent variable and with that it also decides whether a high value means e.g., happiness or unhappiness. If the indicator with the fixed path is scored such that large values indicated happiness and low values unhappiness, the latent variable will be scored accordingly. While the metric disappears in the standardized solution, the direction of the scaling remains consequential.

Don't forget to name all variables (observed and unobserved, including the errors).

What you need to hand in is the print out of the diagram with the standardized coefficients. Using the usual asterisk notation, indicate which paths are significant at the .05, .01 or .001 level. (You may use the infinity row of the t table in your book and treat C.R. as t-values.) Also hand in the overall fit of the model (Chi-square and its degree of freedom).

Answer the following questions:

  1. What is the effect of gender on happiness controlling for individualism?
  2. What is the effect of individualism on happiness controlling for gender?
  3. How well are we explaining why people are happy?
  4. Comment on why you think we have found these results?
  5. What are the strongest indicators of happiness? What are the strongest indicators of individualism? Does that make sense?
  6. How good is the overall fit?