Index construction

[Chiremba]

I have  a quick question.

I would like to create  an index  for decisionmaking in household.
I have  done factor  analysis on a  set  of items  that measure
decisionmaking. Everything is fine  for now.
I'm having a hard  time intepreting my index . Can someone help please.
Detailed  description

Items in questionnaire
q6191 - What type of food to buy for the home
q6192- How much money to spend on food
q6193- When to buy snacks for the kids
q6197- How much money to spend on your children’s clothes
q6198- How much money to spend on your children’s education
q6199- When to spend money for your children’s health

responses
1=Respondent (woman); 2=Husband; 3=Husband and respondent together; 4=Other;
5=No children; 6=Never used money for this purpose

I am interested in  creating a  composite index of  women's participation in
decision making. If response is 1 or  3 then  woman participates  in
decisionmaking. If its 2 then woman does  not participate  in decisionmaking
else  will be treated  as  systems  missing.
The total score  i can get from sum of all factors  is 6 and the lowest 0. 0
to me  means  woman fully participates  in decisionmaking and  a 6  means
she does not at all. However  i do not know how  to interpret values
falling inbetween. Should  I recode all item responses 0 and 1   giving 0
values  if response is 1 and 3 and  giving 1 if response is 2 else   treat
as  systems missing? How  should i intepret this binary variable since
scores  will range from 0 -6.
What procedures  should i use in spss  to  work through this problem. If
anyone can suggest systax that will be helpful.

Yours ,

Chiremba

[zstatman]

Chiremba, this could be straight forward or very complex so, while your description is clear, I would really need to see the data. Thought it might not meet your ultimate needs, there is a paper at my Web site that has an index component: Supplier Side Dynamics case study

Feel free to contact me directly
W
Statistical Services

=========================================
mailto:[hidden email]
http://home.earthlink.net/~statmanz
=========================================

Virus Scan Notice: This email is certified to be virus free.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Chiremba
Sent: Monday, April 30, 2007 8:36 PM
To: [hidden email]
Subject: Index construction

I have  a quick question.

I would like to create  an index  for decisionmaking in household.
I have  done factor  analysis on a  set  of items  that measure decisionmaking. Everything is fine  for now.
I'm having a hard  time intepreting my index . Can someone help please.
Detailed  description

Items in questionnaire
q6191 - What type of food to buy for the home
q6192- How much money to spend on food
q6193- When to buy snacks for the kids
q6197- How much money to spend on your children’s clothes
q6198- How much money to spend on your children’s education
q6199- When to spend money for your children’s health

responses
1=Respondent (woman); 2=Husband; 3=Husband and respondent together; 4=Other; 5=No children; 6=Never used money for this purpose

I am interested in  creating a  composite index of  women's participation in decision making. If response is 1 or  3 then  woman participates  in decisionmaking. If its 2 then woman does  not participate  in decisionmaking else  will be treated  as  systems  missing.
The total score  i can get from sum of all factors  is 6 and the lowest 0. 0 to me  means  woman fully participates  in decisionmaking and  a 6  means she does not at all. However  i do not know how  to interpret values
falling inbetween. Should  I recode all item responses 0 and 1   giving 0
values  if response is 1 and 3 and  giving 1 if response is 2 else   treat
as  systems missing? How  should i intepret this binary variable since scores  will range from 0 -6.
What procedures  should i use in spss  to  work through this problem. If anyone can suggest systax that will be helpful.

Yours ,

Chiremba

[Matthias Spörrle]

Chiremba,

just some logical thoughts concerining your items/procedure:
- If "other" (= 4) means "other person decides": Responses 2 and 4 both
indicate that respondent does not participate in decision. I do not
understand why you should treat 4 as missing since this response indicates
that the woman is not participating (like in 2)
- Note that your coding/responses are not completely crossed: There is no
catagory for "other and respondent"-decision. Can you obtain this from the
data? If not (i.e., the responses were given in close response format) it
reduces the validity of your results, since repondents with this setting (
i.e., decision process between her and other person [e.g., mother]) will
allocate themselves to one of the categories (i.e., 1 or 4) by discounting
and satisficing.
- you can then form indices of (a) how many decisions per person altogether
[excluding missings 5, 6], (b) how many of them were exclusively made by
respondent [1], (c) how many were made in interaction [3 and potentially
recoded 4], (d) how many were made by others [2 and 4]. Using (a) as a
reference catagory will allow you to form rough quantitative estimates of
decision autonomy and interaction.

HTH
Cheers
Matthias




<[hidden email]>

On 5/1/07, Chiremba < [hidden email]> wrote:

>
> I have  a quick question.
>
> I would like to create  an index  for decisionmaking in household.
> I have  done factor  analysis on a  set  of items  that measure
> decisionmaking. Everything is fine  for now.
> I'm having a hard  time intepreting my index . Can someone help please.
> Detailed  description
>
> Items in questionnaire
> q6191 - What type of food to buy for the home
> q6192- How much money to spend on food
> q6193- When to buy snacks for the kids
> q6197- How much money to spend on your children’s clothes
> q6198- How much money to spend on your children’s education
> q6199- When to spend money for your children’s health
>
> responses
> 1=Respondent (woman); 2=Husband; 3=Husband and respondent together;
> 4=Other;
> 5=No children; 6=Never used money for this purpose
>
> I am interested in  creating a  composite index of  women's participation
> in
> decision making. If response is 1 or  3 then  woman participates  in
> decisionmaking. If its 2 then woman does  not participate  in
> decisionmaking
> else  will be treated  as  systems  missing.
> The total score  i can get from sum of all factors  is 6 and the lowest 0.
> 0
> to me  means  woman fully participates  in decisionmaking and  a 6  means
> she does not at all. However  i do not know how  to interpret values
> falling inbetween. Should  I recode all item responses 0 and 1   giving 0
> values  if response is 1 and 3 and  giving 1 if response is 2 else   treat
> as  systems missing? How  should i intepret this binary variable since
> scores  will range from 0 -6.
> What procedures  should i use in spss  to  work through this problem. If
> anyone can suggest systax that will be helpful.
>
> Yours ,
>
> Chiremba
>

[Matthias Spörrle]

Dear Zivai/Chiremba,

I am glad, that you find my comments helpful.

Please send your responses always to the forum. There are many others online
to answer your questions.

I am not sure, yet, what to recommend for your index.
Two thoughts:
- I think it is unwise to drop the category 3, it is very important
information.
- You should form two indices: one for autonomy and one for interaction in
decision. I would guess that they would correlate positively (when
standardized by means of the overall number of decisions per person)

Cheers
Matthias



On 5/1/07, zivai murira <[hidden email]> wrote:

>
> tMany hanks  Matthias.  I appreciate  your  assistance. I think  I will
> proceed  to include  other (4) in  the category where  respondent does  not
> participate in decision-making. Is it appropriate then to proceed to
> transform into binary variables  all the items  I will include  in  index,
> assigning  0  if its  respondent and  2 and  4 assigning  to 1 and  else
> assigning system missing.
> The next challenge  I have is that I  will have  to make  an abitrary
> decision in transforming my index  into a  binary  variable  with 2
> categories  higher  role in  decionmaking and  lower role in decision
> making. How  can I do this  for  example if  total score  is  say 6 and   a
> score of  0 to me  will indicate  highest participation by respondent in
> decision making and  on the other  hand  a score of say 6 means  they do not
> participate at all. My challenge  is intepreting or  categorising  scores in
> the middle ground, 3, 4 etc
>
> Zivai
>
>
> On 5/1/07, Matthias Spörrle <[hidden email]> wrote:
> >
> > Chiremba,
> >
> > just some logical thoughts concerining your items/procedure:
> > - If "other" (= 4) means "other person decides": Responses 2 and 4 both
> > indicate that respondent does not participate in decision. I do not
> > understand why you should treat 4 as missing since this response indicates
> > that the woman is not participating (like in 2)
> > - Note that your coding/responses are not completely crossed: There is
> > no catagory for "other and respondent"-decision. Can you obtain this from
> > the data? If not (i.e., the responses were given in close response
> > format) it reduces the validity of your results, since repondents with this
> > setting ( i.e., decision process between her and other person [e.g.,
> > mother]) will allocate themselves to one of the categories (i.e., 1 or
> > 4) by discounting and satisficing.
> > - you can then form indices of (a) how many decisions per person
> > altogether [excluding missings 5, 6], (b) how many of them were exclusively
> > made by respondent [1], (c) how many were made in interaction [3 and
> > potentially recoded 4], (d) how many were made by others [2 and 4]. Using
> > (a) as a reference catagory will allow you to form rough quantitative
> > estimates of decision autonomy and interaction.
> >
> > HTH
> > Cheers
> > Matthias
> >
> >
> >
> >
> > <[hidden email]>
> >
> > On 5/1/07, Chiremba < [hidden email]> wrote:
> > >
> > > I have  a quick question.
> > >
> > > I would like to create  an index  for decisionmaking in household.
> > > I have  done factor  analysis on a  set  of items  that measure
> > > decisionmaking. Everything is fine  for now.
> > > I'm having a hard  time intepreting my index . Can someone help
> > > please.
> > > Detailed  description
> > >
> > > Items in questionnaire
> > > q6191 - What type of food to buy for the home
> > > q6192- How much money to spend on food
> > > q6193- When to buy snacks for the kids
> > > q6197- How much money to spend on your children’s clothes
> > > q6198- How much money to spend on your children’s education
> > > q6199- When to spend money for your children’s health
> > >
> > > responses
> > > 1=Respondent (woman); 2=Husband; 3=Husband and respondent together;
> > > 4=Other;
> > > 5=No children; 6=Never used money for this purpose
> > >
> > > I am interested in  creating a  composite index of  women's
> > > participation in
> > > decision making. If response is 1 or  3 then  woman participates  in
> > > decisionmaking. If its 2 then woman does  not participate  in
> > > decisionmaking
> > > else  will be treated  as  systems  missing.
> > > The total score  i can get from sum of all factors  is 6 and the
> > > lowest 0. 0
> > > to me  means  woman fully participates  in decisionmaking and  a
> > > 6  means
> > > she does not at all. However  i do not know how  to interpret values
> > > falling inbetween. Should  I recode all item responses 0 and 1
> > > giving 0
> > > values  if response is 1 and 3 and  giving 1 if response is 2 else
> > > treat
> > > as  systems missing? How  should i intepret this binary variable since
> > >
> > > scores  will range from 0 -6.
> > > What procedures  should i use in spss  to  work through this problem.
> > > If
> > > anyone can suggest systax that will be helpful.
> > >
> > > Yours ,
> > >
> > > Chiremba
> > >
> >
> >
>

[Matthias Spörrle]

Dear Zivai/Chiremba,

I am glad, that you find my comments helpful.

Please send your responses always to the forum. There are many others online
to answer your questions.

I am not sure, yet, what to recommend for your index.
Two thoughts:
- I think it is unwise to drop the category 3, it is very important
information.
- You should form two indices: one for autonomy and one for interaction in
decision. I would guess that they would correlate positively (when
standardized by means of the overall number of decisions per person)

Cheers
Matthias



On 5/1/07, zivai murira <[hidden email]> wrote:

>
> tMany hanks  Matthias.  I appreciate  your  assistance. I think  I will
> proceed  to include  other (4) in  the category where  respondent does  not
> participate in decision-making. Is it appropriate then to proceed to
> transform into binary variables  all the items  I will include  in  index,
> assigning  0  if its  respondent and  2 and  4 assigning  to 1 and  else
> assigning system missing.
> The next challenge  I have is that I  will have  to make  an abitrary
> decision in transforming my index  into a  binary  variable  with 2
> categories  higher  role in  decionmaking and  lower role in decision
> making. How  can I do this  for  example if  total score  is  say 6 and   a
> score of  0 to me  will indicate  highest participation by respondent in
> decision making and  on the other  hand  a score of say 6 means  they do not
> participate at all. My challenge  is intepreting or  categorising  scores in
> the middle ground, 3, 4 etc
>
> Zivai
>
>
> On 5/1/07, Matthias Spörrle < [hidden email]> wrote:
> >
> > Chiremba,
> >
> > just some logical thoughts concerining your items/procedure:
> > - If "other" (= 4) means "other person decides": Responses 2 and 4 both
> > indicate that respondent does not participate in decision. I do not
> > understand why you should treat 4 as missing since this response indicates
> > that the woman is not participating (like in 2)
> > - Note that your coding/responses are not completely crossed: There is
> > no catagory for "other and respondent"-decision. Can you obtain this from
> > the data? If not (i.e., the responses were given in close response
> > format) it reduces the validity of your results, since repondents with this
> > setting ( i.e., decision process between her and other person [e.g.,
> > mother]) will allocate themselves to one of the categories (i.e., 1 or
> > 4) by discounting and satisficing.
> > - you can then form indices of (a) how many decisions per person
> > altogether [excluding missings 5, 6], (b) how many of them were exclusively
> > made by respondent [1], (c) how many were made in interaction [3 and
> > potentially recoded 4], (d) how many were made by others [2 and 4]. Using
> > (a) as a reference catagory will allow you to form rough quantitative
> > estimates of decision autonomy and interaction.
> >
> > HTH
> > Cheers
> > Matthias
> >
> >
> >
> >
> > <[hidden email]>
> >
> > On 5/1/07, Chiremba < [hidden email]> wrote:
> > >
> > > I have  a quick question.
> > >
> > > I would like to create  an index  for decisionmaking in household.
> > > I have  done factor  analysis on a  set  of items  that measure
> > > decisionmaking. Everything is fine  for now.
> > > I'm having a hard  time intepreting my index . Can someone help
> > > please.
> > > Detailed  description
> > >
> > > Items in questionnaire
> > > q6191 - What type of food to buy for the home
> > > q6192- How much money to spend on food
> > > q6193- When to buy snacks for the kids
> > > q6197- How much money to spend on your children’s clothes
> > > q6198- How much money to spend on your children’s education
> > > q6199- When to spend money for your children’s health
> > >
> > > responses
> > > 1=Respondent (woman); 2=Husband; 3=Husband and respondent together;
> > > 4=Other;
> > > 5=No children; 6=Never used money for this purpose
> > >
> > > I am interested in  creating a  composite index of  women's
> > > participation in
> > > decision making. If response is 1 or  3 then  woman participates  in
> > > decisionmaking. If its 2 then woman does  not participate  in
> > > decisionmaking
> > > else  will be treated  as  systems  missing.
> > > The total score  i can get from sum of all factors  is 6 and the
> > > lowest 0. 0
> > > to me  means  woman fully participates  in decisionmaking and  a
> > > 6  means
> > > she does not at all. However  i do not know how  to interpret values
> > > falling inbetween. Should  I recode all item responses 0 and 1
> > > giving 0
> > > values  if response is 1 and 3 and  giving 1 if response is 2 else
> > > treat
> > > as  systems missing? How  should i intepret this binary variable since
> > >
> > > scores  will range from 0 -6.
> > > What procedures  should i use in spss  to  work through this problem.
> > > If
> > > anyone can suggest systax that will be helpful.
> > >
> > > Yours ,
> > >
> > > Chiremba
> > >
> >
> >
>

[Hector Maletta]

Dear Chiremba,
while the discussion by Mathis Sporrie about the response categories is interesting and useful, I wish to go back to your original question, accepting as given your questions and response conventions.
The application of factor analysis implies that you believe the whole issue of women participation in decision making about children feeding is governed by one or more underlying, unobserved, continuous variable/s. Responses to your questions merely reflect the different positions of respondents along these underlying variables.
Now, I wonder how did you do such factor analysis. You ask whether you should recode your responses to 0, 1 and system missing, implying you have not yet done so. If that is true, then either (1) you have treated the conventional codes of your categorical responses as interval values, which is incorrect, or (2) you have applied categorical factor analysis (procedure CATCPA) which is unlikely since you do not mention it. You COULD have performed a classical factor analysis after recoding your responses to 0-1; it is often done, though some objections exist to this procedure (the main one being that residuals of regressions, on which factor analysis is based, are not normally distributed). However, many practitioners think these objections are not significant in many cases, and do it anyway.
Even if you do not give many details, I assume your "index" is the factor score for the first factor extracted from your items. This score is a measure of the underlying variable, with mean zero and SD=1, ranging from minimum to maximum participation (or the converse, depending on what is 0 and what is 1 in your variables). The interpretation is easy: the higher the index, the higher (or lower) the degree of women participation in decision making about food given to children.
Your dichotomous items (responses reduced to 0-1) can be seen as gauges of the underlying value. For each subject, the probability of getting a value of 1 would be a function of the value of that subject in the underlying unobserved variable.

Now, you might also build your index in a more simple way, such as simply adding one point for each positive response. This is undoubtedly crude (all items are given the same weight) but can be handy for your purposes. The RELIABILITY procedure in SPSS can give you an analysis about whether your items add up adequately to such an additive scale.

Hector

----- Mensaje original -----
De: Chiremba <[hidden email]>
Fecha: Martes, Mayo 1, 2007 2:36 am
Asunto: Index construction

> I have  a quick question.
>
> I would like to create  an index  for decisionmaking in household.
> I have  done factor  analysis on a  set  of items  that measure
> decisionmaking. Everything is fine  for now.
> I'm having a hard  time intepreting my index . Can someone help
> please.Detailed  description
>
> Items in questionnaire
> q6191 - What type of food to buy for the home
> q6192- How much money to spend on food
> q6193- When to buy snacks for the kids
> q6197- How much money to spend on your children’s clothes
> q6198- How much money to spend on your children’s education
> q6199- When to spend money for your children’s health
>
> responses
> 1=Respondent (woman); 2=Husband; 3=Husband and respondent
> together; 4=Other;
> 5=No children; 6=Never used money for this purpose
>
> I am interested in  creating a  composite index of  women's
> participation in
> decision making. If response is 1 or  3 then  woman participates  in
> decisionmaking. If its 2 then woman does  not participate  in
> decisionmakingelse  will be treated  as  systems  missing.
> The total score  i can get from sum of all factors  is 6 and the
> lowest 0. 0
> to me  means  woman fully participates  in decisionmaking and  a 6
> means
> she does not at all. However  i do not know how  to interpret values
> falling inbetween. Should  I recode all item responses 0 and 1
> giving 0
> values  if response is 1 and 3 and  giving 1 if response is 2 else
>  treat
> as  systems missing? How  should i intepret this binary variable since
> scores  will range from 0 -6.
> What procedures  should i use in spss  to  work through this
> problem. If
> anyone can suggest systax that will be helpful.
>
> Yours ,
>
> Chiremba
>

[Richard Ristow]

At 03:08 AM 5/2/2007, Hector Maletta wrote:

>I wish to go back to your original question, accepting as given your
>questions and response conventions.
>The application of factor analysis implies that you believe the whole
>issue [...] is governed by one or more underlying, unobserved,
>continuous variable/s. [...]

This is a first-rate response - very clear, very informative. Thank
you!
Richard