testing differences changes

[Antoon Smulders]

 

 

Hello all,

 

My datafile (from an external source) is essentially  structured as a (series of) crosstabulations. A simplified example:

 

Year    Region  Gender  Age     Total

2008    All     male    12-18   xxx

2008    All     female  12-18   xxx

2008    All     male    18-24   xxx

2008    All     female  18-24   xxx

2008    North   male    12-18   xxx

2008    East    male    12-18   xxx

….

2008    North   male    18-24   xxx

…

2015    All     male    12-18   xxx

Etc.

 

The “Total” variable indicates the total number of crime suspects for the value combination of the other variables.

 

I know how to use SELECT IF and WEIGHT to produce some CROSSTABS and the significances.

For instance:

 

WEIGHT BY Total.

SELECT IF Region = “All”.

CROSSTABS TABLES Age BY Year /* there are only two years in this example */ BY Gender /statistics chisq phi.

 

This code gives me two significant p-values indicating (for instance) that the total number decreased for as well male as female suspects.

 

Now my question: is it possible to test the hypothesis that the decrease for males is significantly different from the decrease for females? My first thought is that I have to test the differences in phi or Cramer’s V, but I don’t know how to do that. Any suggestions?

 

 

 

 

 

===================== 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

[Maguin, Eugene]

Why not

Crosstabs age by gender/….

 

Gene Maguin

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Antoon Smulders
Sent: Monday, November 07, 2016 6:04 AM
To: [hidden email]
Subject: testing differences changes

 

 

 

Hello all,

 

My datafile (from an external source) is essentially  structured as a (series of) crosstabulations. A simplified example:

 

Year    Region  Gender  Age     Total

2008    All     male    12-18   xxx

2008    All     female  12-18   xxx

2008    All     male    18-24   xxx

2008    All     female  18-24   xxx

2008    North   male    12-18   xxx

2008    East    male    12-18   xxx

….

2008    North   male    18-24   xxx

…

2015    All     male    12-18   xxx

Etc.

 

The “Total” variable indicates the total number of crime suspects for the value combination of the other variables.

 

I know how to use SELECT IF and WEIGHT to produce some CROSSTABS and the significances.

For instance:

 

WEIGHT BY Total.

SELECT IF Region = “All”.

CROSSTABS TABLES Age BY Year /* there are only two years in this example */ BY Gender /statistics chisq phi.

 

This code gives me two significant p-values indicating (for instance) that the total number decreased for as well male as female suspects.

 

Now my question: is it possible to test the hypothesis that the decrease for males is significantly different from the decrease for females? My first thought is that I have to test the differences in phi or Cramer’s V, but I don’t know how to do that. Any suggestions?

 

 

 

 

 

===================== 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

[Bruce Weaver]

You ask, "is it possible to test the hypothesis that the decrease for males is significantly different from the decrease for females?"

If your research questions concern how Year, Age & Gender influence the total number of crime suspects, I don't think you should be using CROSSTABS at all.  Your outcome is not one of those categorical variables.  Rather, it is the number of crime suspects.  So I think you should be using some kind of regression model.  On the face of it, it might look like a Poisson (or negative binomial) model, given that the outcome is a count.  But if the xxx values below are fairly large (which I imagine they are), a linear model might work just as well (if not better).  (Whether it should be OLS regression or a multilevel model is not entirely clear.)    

If the 'decrease' you are talking about in your question above refers to a drop in the numbers over time, then you need to look at the interaction between Year and Gender.  

HTH.

Antoon Smulders wrote

Hello all,

My datafile (from an external source) is essentially  structured as a (series of) crosstabulations. A simplified example:

Year    Region  Gender  Age     Total
2008    All     male    12-18   xxx
2008    All     female  12-18   xxx
2008    All     male    18-24   xxx
2008    All     female  18-24   xxx
2008    North   male    12-18   xxx
2008    East    male    12-18   xxx
....
2008    North   male    18-24   xxx
...
2015    All     male    12-18   xxx
Etc.

The "Total" variable indicates the total number of crime suspects for the value combination of the other variables.

I know how to use SELECT IF and WEIGHT to produce some CROSSTABS and the significances.
For instance:

WEIGHT BY Total.
SELECT IF Region = "All".
CROSSTABS TABLES Age BY Year /* there are only two years in this example */ BY Gender /statistics chisq phi.

This code gives me two significant p-values indicating (for instance) that the total number decreased for as well male as female suspects.

Now my question: is it possible to test the hypothesis that the decrease for males is significantly different from the decrease for females? My first thought is that I have to test the differences in phi or Cramer's V, but I don't know how to do that. Any suggestions?






=====================
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

[David Greenberg]

Cross-tabs is not the way to analyze panel data with a continuous
outcome. You should be thinking of fixed-effects  panel regression
methods. SPSS is not well-equipped for carrying out such econometric
methods. David Greenberg, Sociology Department, New York U.

On Mon, Nov 7, 2016 at 6:04 AM, Antoon Smulders <[hidden email]> wrote:

>
>
> Hello all,
>
> My datafile (from an external source) is essentially  structured as a
> (series of) crosstabulations. A simplified example:
>
> Year    Region  Gender  Age     Total
> 2008    All     male    12-18   xxx
> 2008    All     female  12-18   xxx
> 2008    All     male    18-24   xxx
> 2008    All     female  18-24   xxx
> 2008    North   male    12-18   xxx
> 2008    East    male    12-18   xxx
> ….
> 2008    North   male    18-24   xxx
> …
> 2015    All     male    12-18   xxx
> Etc.
>
> The “Total” variable indicates the total number of crime suspects for the
> value combination of the other variables.
>
> I know how to use SELECT IF and WEIGHT to produce some CROSSTABS and the
> significances.
> For instance:
>
> WEIGHT BY Total.
> SELECT IF Region = “All”.
> CROSSTABS TABLES Age BY Year /* there are only two years in this example */
> BY Gender /statistics chisq phi.
>
> This code gives me two significant p-values indicating (for instance) that
> the total number decreased for as well male as female suspects.
>
> Now my question: is it possible to test the hypothesis that the decrease for
> males is significantly different from the decrease for females? My first
> thought is that I have to test the differences in phi or Cramer’s V, but I
> don’t know how to do that. Any suggestions?
>
>
>
>
>
> ===================== 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

[Mike]

I think that there is a fundamental confusion about what analysis
the OP wants to do because the crosstabs example provided
below just looks at the categorical variables and does not involve
"Total" which presumably is a continuous variable.  Bruce Weaver
had focused on how to analyze the data if the intent is to related
the categorical variables to the "Total" variable -- though it is not
clear that this is what the OP wants to do.  With respect to David
Greenberg, he seems to be in agreement with Bruce but again,
it is not clear that the OP wants to use Total in the analysis (at
least not yet).

With respect to "fixed effects panel regression", though Stata
seems better suited for this type of analysis, SPSS may handle
the situation if one conceives of the design as a multilevel
regression problem.  One source that I would suggest the
following reference:

Brüderl, J. & Ludwig, V. (2015). Fixed-effects panel regression.
In H. Best & C. Wolf The SAGE handbook of regression analysis
and causal inference (pp. 327-358). London: SAGE Publications Ltd.
doi: 10.4135/9781446288146.n15

Some of the other chapters may also be relevant.  However,
I do think that the OP needs to be clearer on what the research
question or goal is of the analysis.  If Total is not involved,
wouldn't log-linear analysis be a more appropriate analysis?
Just saying.

-Mike Palij
New York University
[hidden email]


----- Original Message -----
From: "David Greenberg" <[hidden email]>
To: <[hidden email]>
Sent: Monday, November 07, 2016 12:18 PM
Subject: Re: testing differences changes


Cross-tabs is not the way to analyze panel data with a continuous
outcome. You should be thinking of fixed-effects  panel regression
methods. SPSS is not well-equipped for carrying out such econometric
methods. David Greenberg, Sociology Department, New York U.

On Mon, Nov 7, 2016 at 6:04 AM, Antoon Smulders <[hidden email]>
wrote:

>
>
> Hello all,
>
> My datafile (from an external source) is essentially  structured as a
> (series of) crosstabulations. A simplified example:
>
> Year    Region  Gender  Age     Total
> 2008    All     male    12-18   xxx
> 2008    All     female  12-18   xxx
> 2008    All     male    18-24   xxx
> 2008    All     female  18-24   xxx
> 2008    North   male    12-18   xxx
> 2008    East    male    12-18   xxx
> ….
> 2008    North   male    18-24   xxx
> …
> 2015    All     male    12-18   xxx
> Etc.
>
> The “Total” variable indicates the total number of crime suspects for
> the
> value combination of the other variables.
>
> I know how to use SELECT IF and WEIGHT to produce some CROSSTABS and
> the
> significances.
> For instance:
>
> WEIGHT BY Total.
> SELECT IF Region = “All”.
> CROSSTABS TABLES Age BY Year /* there are only two years in this
> example */
> BY Gender /statistics chisq phi.
>
> This code gives me two significant p-values indicating (for instance)
> that
> the total number decreased for as well male as female suspects.
>
> Now my question: is it possible to test the hypothesis that the
> decrease for
> males is significantly different from the decrease for females? My
> first
> thought is that I have to test the differences in phi or Cramer’s V,
> but I
> don’t know how to do that. Any suggestions?
>
>
>
>
>
> ===================== 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

=====================
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