SPSSX Discussion

lag function

Classic

List

Threaded

10 messages Options

Keval Khichadia

lag function

Hi,

ID       TERM                  CUMGPA             USEGPA

1       FALL08                        3.5                       3.5
1       SUMMER08                3.4                        3.5
1       SPRING08                    3.3                        3.5
2        FALL08                                                    3.9
2        SPRING08                    3.9                        3.9
2        FALL07                        3.8                        3.9
3        FALL08                                                    3.1
3        SPRING08                                                3.1
3        FALL07                        3.1                        3.1

In the example above, I need to select the GPA for the student from the latest term we have data for that student and write that value into a new column (USEGPA) for all cases of each unique student.

Thanks for any help,

Keval

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

Nancy Darling-2

Re: lag function

You can
1) recode the terms (currently string functions) to numerical numbered
from oldest to most recent.
2) sort by id term.
3) create an aggregate variable that identifies the value of the last term
AGGREGATE
/OUTFILE=*
MODE=ADDVARIABLES
/BREAK=id
/recentgpa_last = last(cumgpa).

That variable should be your most recent gpa. If you don't want the
variable in your current dataset, you can have aggregate export as a
separate dataset and merge it later.

Keval Khichadia wrote:

> Hi,
>
>
> ID TERM CUMGPA USEGPA
>
> 1 FALL08 3.5 3.5
> 1 SUMMER08 3.4 3.5
> 1 SPRING08 3.3 3.5
> 2 FALL08 3.9
> 2 SPRING08 3.9 3.9
> 2 FALL07 3.8 3.9
> 3 FALL08 3.1
> 3 SPRING08 3.1
> 3 FALL07 3.1 3.1
>
> In the example above, I need to select the GPA for the student from the latest term we have data for that student and write that value into a new column (USEGPA) for all cases of each unique student.
>
>
> Thanks for any help,
>
> Keval
>
>
>
>
> ===================
> 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

Art Kendall

Re: lag function

In reply to this post by Keval Khichadia

Start a new instance of SPSS. Copy, paste, and run this syntax.
Does this do what you want?
Note that the LAST function gives the last legitimate value.

Also if you are not the creator of this data be careful about what the
year means. Sometimes a school term that starts in the Fall of 08 and
end in Spring 09 is called 08 and sometimes is is called 09 as in
Class of 09.

data list list/
ID(f1) TERM(a6) Yr(n2) CUMGPA(f4.1) USEGPA(f4.1).
begin data
1 FALL 08 3.5 3.5
1 SUMMER 08 3.4 3.5
1 SPRING 08 3.3 3.5
2 FALL 08 -1 3.9
2 SPRING 08 3.9 3.9
2 FALL 07 3.8 3.9
3 FALL 08 -1 3.1
3 SPRING 08 -1 3.1
3 FALL 07 3.1 3.1
end data.
missing values cumgpa(-1).
FORMATS YR(N2).
recode term ('SPRING'=1)('SUMMER'=2)('FALL'=3) INTO NTERM.
SORT CASES BY ID YR NTERM.
AGGREGATE OUTFILE=* MODE= ADDVARIABLES OVERWRITE=YES
/BREAK=id
/LASTGPA = LAST(CUMGPA).
LIST.

Art Kendall
Social Research Consultants
Keval Khichadia wrote:

Art Kendall
Social Research Consultants

Gregor Selčan

which test?

In reply to this post by Nancy Darling-2

Hi...

i have following variables:
mean1 (scale variable)
mean2 (scale variable); mean2 << mean1
age_groups (ordinal variable; 5 variable values)

it is quite simple to find out if there is any significant difference
between mean1 or mean2 by age_groups. ANOVA will be quite appropriate
test... but...

how should I resolve this problem: is there any significant difference
between mean1 and mean2 within each of age_group?

Thanks in advance,

Gregor

__________ Information from ESET NOD32 Antivirus, version of virus signature
database 3632 (20081121) __________

The message was checked by ESET NOD32 Antivirus.

http://www.eset.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

Art Kendall

Re: which test?

Please give us a little more detail.
Do you only have data that is already aggregated to the means? Or do you
have the raw data?
Did the age come in the 5 groups of do you have the actual ages?

Are mean1 and mean2 some form of repeated measurement? pre and post?
members of a couple? the same subjects under different treatments?

Did you have a significant repeated variable * age group interaction in
the ANOVA and want to know about post-hoc tests?

Art Kendall
Social Research Consultants

Gregor Selčan wrote:

> Hi...
>
> i have following variables:
> mean1 (scale variable)
> mean2 (scale variable); mean2 << mean1
> age_groups (ordinal variable; 5 variable values)
>
> it is quite simple to find out if there is any significant difference
> between mean1 or mean2 by age_groups. ANOVA will be quite appropriate
> test... but...
>
> how should I resolve this problem: is there any significant difference
> between mean1 and mean2 within each of age_group?
>
> Thanks in advance,
>
> Gregor
>
>
>
> __________ Information from ESET NOD32 Antivirus, version of virus signature
> database 3632 (20081121) __________
>
> The message was checked by ESET NOD32 Antivirus.
>
> http://www.eset.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
>
>
>

Art Kendall
Social Research Consultants

Gregor Selčan

Re: which test?

Thank for your answers!

@Art: I have raw data (n=500). Respondents have had to evaluate two
statements using 5-scale of agreement: Mean1 and Mean2 are not form of
repeated measurements.

I also have actual age which I recoded into 5 groups. Subjects are under
same treatments.

I'd like to know how to find out if there is any significant difference
between mean1 and mean2 within each age group? I am not sure Turkey HSD test
is appropriate.

Thanks,

G.

__________ Information from ESET NOD32 Antivirus, version of virus signature
database 3632 (20081121) __________

The message was checked by ESET NOD32 Antivirus.

http://www.eset.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

Art Kendall

Re: which test?

It still sounds as if the two means are within-subject (repeated
measurement under some change in circumstance).
If that is so here are two ways to control for age.
One by leaving it as a continuous, and one by breaking it into groups.

If the means you are interested in comparing are based on a between
subject factor. Please post that info back.

The syntax below makes up some data that is how I am interpreting your
query.
Open a new instance of SPSS. Copy, Paste, and Run this syntax.
Does the data look similar to your data?

There are 3 kinds of post-hoc test requested. The choice of what kind
of post-hoc to use becomes moot

* make up some data.
new file.
set seed = 20081123.
input program.
loop id = 1 to 500.
compute age = rnd(rv.uniform(17.5, 75.5)).
compute score1 = rnd(rv.uniform(.5, 5.5)).
compute score2 = rnd(rv.uniform(.5, 5.5)).
end case.
end loop.
end file.
end input program.
formats id (f3) age (f2) score1 score2 (f1).
rank /variables=age /ntiles(5).

*control for age by using it as a covariate.
GLM score1 score2 WITH age
/WSFACTOR=score 2 Repeated
/METHOD=SSTYPE(3)
/EMMEANS=TABLES(OVERALL) WITH(age=MEAN)
/EMMEANS=TABLES(score) WITH(age=MEAN)
/CRITERIA=ALPHA(.05)
/WSDESIGN=score
/DESIGN=age.
*control for age by using it as ordered groups.
GLM score1 score2 BY Nage
/WSFACTOR=score 2 Repeated
/CONTRAST(Nage)=Polynomial
/METHOD=SSTYPE(3)
/POSTHOC=Nage(SCHEFFE LSD T2)
/PLOT=PROFILE(Nage*score)
/EMMEANS=TABLES(OVERALL)
/EMMEANS=TABLES(score)
/CRITERIA=ALPHA(.05)
/WSDESIGN=score
/DESIGN=Nage.

Art Kendall
Social Research Consultants

Gregor Selčan wrote:

> Thank for your answers!
>
> @Art: I have raw data (n=500). Respondents have had to evaluate two
> statements using 5-scale of agreement: Mean1 and Mean2 are not form of
> repeated measurements.
>
> I also have actual age which I recoded into 5 groups. Subjects are under
> same treatments.
>
> I'd like to know how to find out if there is any significant difference
> between mean1 and mean2 within each age group? I am not sure Turkey HSD test
> is appropriate.
>
> Thanks,
>
> G.
>
>
>
> __________ Information from ESET NOD32 Antivirus, version of virus signature
> database 3632 (20081121) __________
>
> The message was checked by ESET NOD32 Antivirus.
>
> http://www.eset.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
>
>
>

Art Kendall
Social Research Consultants

Kreischer,Resha M

data type

I am going to do some analyses on Final Grades (A, B, C, D, F). Should I
consider Final Grades as ordinal data? Or, if I recode A to 4 points, B
to 3 points, C to 2 points, D to 1 point and F to 0 points, would it be
considered interval/ratio?

Thanks for your feedback.

Resha

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

Re: data type

Resha,

What you do depends how you plan analyze the data and what your needs are
for their presentation. In any event, I'd recode your data to numbers to
give yourself more analytical flexibility. If your audience expects to see
GPAs, then you need 'continuous' data type analyses. If the interest is in
letter grade distribution, then categorical type analyses. But underlying
those two comments, do you think that the increment in subject
competency/knowledge as you move from F to D to C to B to A is constant? I
don't. Do you think that a C indicates half the subject competency/knowledge
as an A? I don't. But, the assumption is a useful fiction.

Gene Maguin

>>I am going to do some analyses on Final Grades (A, B, C, D, F). Should I
consider Final Grades as ordinal data? Or, if I recode A to 4 points, B
to 3 points, C to 2 points, D to 1 point and F to 0 points, would it be
considered interval/ratio?

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

Oliver, Richard

Re: data type

In reply to this post by Kreischer,Resha M

Changing letters to numbers doesn't change the measurement level. Numeric codes are frequently used for nominal and ordinal data (e.g. 1=Strongly Agree, 2=Agree, etc.). It's doubtful that you would want to treat the data as ratio. Would you consider an A to be twice as good as a C or 4 times as good as a D? Treating the data as interval has similar issues. Is the distance between an A and a B the same as the distance between a C and a D?

That said, I think that survey researchers frequently treat Likert scale ordinal data as interval for the purpose of computing means, although I'd be more cautious in using statistical tests that rely on certain assumptions, such as normally distributed data.

I'm sure those with more statistical knowledge have better informed opinions on the subject.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Kreischer,Resha M
Sent: Wednesday, December 03, 2008 8:55 AM
To: [hidden email]
Subject: data type

I am going to do some analyses on Final Grades (A, B, C, D, F). Should I
consider Final Grades as ordinal data? Or, if I recode A to 4 points, B
to 3 points, C to 2 points, D to 1 point and F to 0 points, would it be
considered interval/ratio?

Thanks for your feedback.

Resha

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