How to analyse characteristics of a population

[zwaluw]

Hi,

It would be great if you could point out what analyses I need to use. Thanks a lot in advance!

I have a data-set of variables on 63 firms (independent variables) and preferences of these firms to collaborate with another firm (dependent variables).

I have completed analyzing the main and moderating effects. But now I'd like to provide more insight that is of interest for practice, for the managers of these firms.


I'd like to analyze:

1a) what the characteristics are of the firms that were taken into the mixed model to measure main effects.

1b) Can you also analyse these characteristics for firms of which the standard error was low (so actual values close to estimated values), and compare them with firms of which str. error was high?

2) what the characteristics are of the firms when you take just one variable and its descriptive statistics. For instance: you take alliance experience as a variable. The mean alliance experience is 4 (4 prior alliances). Then I'd like to know: looking at all these that on average had 4 prior alliances, how much do these firms spend on research and development (R&D)? It would be ideal to see a sort of visualization showing, for instance, that around the average of 4 alliances, firms were spending rather much on R&D, while firms with many prior alliances, were not spending much on R&D.

Thanks again, hope you can help!

[Bruce Weaver]

This looks like a homework assignment.  If it is...

What have you tried so far?  Have you asked your instructor or teaching assistant for guidance?

zwaluw wrote

Hi,

It would be great if you could point out what analyses I need to use. Thanks a lot in advance!

I have a data-set of variables on 63 firms (independent variables) and preferences of these firms to collaborate with another firm (dependent variables).

I have completed analyzing the main and moderating effects. But now I'd like to provide more insight that is of interest for practice, for the managers of these firms.


I'd like to analyze:

1a) what the characteristics are of the firms that were taken into the mixed model to measure main effects.

1b) Can you also analyse these characteristics for firms of which the standard error was low (so actual values close to estimated values), and compare them with firms of which str. error was high?

2) what the characteristics are of the firms when you take just one variable and its descriptive statistics. For instance: you take alliance experience as a variable. The mean alliance experience is 4 (4 prior alliances). Then I'd like to know: looking at all these that on average had 4 prior alliances, how much do these firms spend on research and development (R&D)? It would be ideal to see a sort of visualization showing, for instance, that around the average of 4 alliances, firms were spending rather much on R&D, while firms with many prior alliances, were not spending much on R&D.

Thanks again, hope you can help!

[David Marso]

To me it sounds like a consulting opportunity ;-)

Bruce Weaver wrote

This looks like a homework assignment.  If it is...

What have you tried so far?  Have you asked your instructor or teaching assistant for guidance?

zwaluw wrote

Hi,

It would be great if you could point out what analyses I need to use. Thanks a lot in advance!

I have a data-set of variables on 63 firms (independent variables) and preferences of these firms to collaborate with another firm (dependent variables).

I have completed analyzing the main and moderating effects. But now I'd like to provide more insight that is of interest for practice, for the managers of these firms.


I'd like to analyze:

1a) what the characteristics are of the firms that were taken into the mixed model to measure main effects.

1b) Can you also analyse these characteristics for firms of which the standard error was low (so actual values close to estimated values), and compare them with firms of which str. error was high?

2) what the characteristics are of the firms when you take just one variable and its descriptive statistics. For instance: you take alliance experience as a variable. The mean alliance experience is 4 (4 prior alliances). Then I'd like to know: looking at all these that on average had 4 prior alliances, how much do these firms spend on research and development (R&D)? It would be ideal to see a sort of visualization showing, for instance, that around the average of 4 alliances, firms were spending rather much on R&D, while firms with many prior alliances, were not spending much on R&D.

Thanks again, hope you can help!

[zwaluw]

Hi Bruce, well I've tried the two-step and K-means clustering to try to classify groups but just not sure what this in the end tells me. Which analysis to use is just what I'm interested in. And it's meant for a government agency, which are sponsoring (my) university research. Didn't know this would be an issue?

[Mike]

On: Friday, July 29, 2016 5:19 AM, zwaluw wrote:

Just off the top of my head, some suggestions.  There probably
are others who can provide better suggestions.

>Subject: How to analyse characteristics of a population

First, you do not identify the total number of "cases" that
constitute your "population".  Do you really mean population,
that is, you are only interested in the N of cases you have
and there is no larger group of firms/companies/whatever
that you are concerned with?  If you are serious that you
have are analyzing a population, you're going to have to
go back and alter some statistics (e.g., standard deviation
should have N instead of N-1 in their denominator).

If you are really analyzing a population, you can't be doing
"inferential statitical tests" because you're not trying to infer
the properties of the population your numbers come from --
you already have the population.  All you can do is use the
appropriate descriptive statistics and test different models
of relationships (say through bootstrapping or permutation
tests).


> Hi,
>
> It would be great if you could point out what analyses I need
>to use. Thanks a lot in advance!
>
> I have a data-set of variables on 63 firms (independent variables)

It may obvious to you because you are familiar with the
dataset but others may have difficulty understand what you
are saying here::

(1) When you say "independent variables" are you referring to
(a) the 63 firms/companies (i.e., a one-way 63 level between-subject
design) or the variables you measured on these 63 units?

(2) What do you mean when you use the term "independent
variables"?  Again, it is unclear what your "independent variables"
are but did you have random assignment of units/cases to the
levels of the independent variables?  If not, the variables are
measures of attributes of the units of analysis (like the sex, age,
SEX, etc., of a person) and are not "independent variables" but,
at best, "quasi-independent variables", that is, variables that
may have a causal effects on other variables (which some might
call dependent variables" but if you're not really using an experimental
design, you should call them "outcome variables" or "effects"
something similar so that the reader is confused into thinking
you are modeling an experimentally derived causal process --
if you think you are modeling a causal process, you need to
explain the rationale for it.

Jon Peck can correct me on the following point, but econometricians
have the distressing habit of calling naturally occruing "causal
variables" as "independent variables", ignoring the fact that
in experimental designs "independent variables" are under
the control of the experimenter while naturally occuring casual
variables (e.g., yearly salary) are not.  Is your use of the term
"independent variables" consistent with traditional experimental
design and analysis or econometric/causal analysis?

>and preferences of these firms to collaborate with another firm
>(dependent  variables).

Again, it is unclear what your variables are but let us assume
you have a one-way 63 level between-subjects design and you
have, say, 10 variables that describe attributes of the 63
firms that form the levels of this design.  These attributes
might be variables that are dichotomies, ordinal (ratings, ranks,
etc.), interval or ratio.  Presumably you have sumarized these
variables for each of the 63 firms in some appropriate way
(this is important for your later questions).

> I have completed analyzing the main and moderating effects.

If you are dealing with a a population, I assume that you've
just done a descriptive analysis of the 63 firms/cases and
tried out different patterns of correlation or covariance matrix
to determine which pattern "fits" best perhaps through some
permutation test since you cannot use inferential tests.

> But now I'd like
> to provide more insight that is of interest for practice, for
> the managers of these firms.

Okay.  But are you trying to describe past behavior or predict
future behavior?  If the latter, how do you assess predictive validity?

> I'd like to analyze:
>
> 1a) what the characteristics are of the firms that were taken into the
> mixed model to measure main effects.

Again, if you used a population, this statement makes no sense.

If you used samples, then
(a) You can provide a table with the descriptive statistics for the
background variables that describe these the firms/cases (?)
used in the whatever analysis you did,
or
(b) You could construct a path diagram that identifies the background
variables that significantly discriminate among the firms/cases AND
which are related to your outcome variable.

> 1b) Can you also analyse these characteristics for firms of which the
> standard error was low (so actual values close to estimated values),
> and
> compare them with firms of which str. error was high?

There are actually *at least* two aspects to the question above:
(a) what are the reliabilities of the measures used (low reliabilities
will increase the standard error), and
(b) if you are talking about standard errors of relationships (e.g.,
correlations), then you're referring to how well you measuring the
relationship between variables (i.e., comparing strong relationships
versus weak relationships).  This will be afftected by the reliability
of
the measures, whether you have adequately controlled for 3rd or
nuisance variables (i.e., you don't have a spurious correlation) and
the magnitude of the relationship in the population (if you are working
from samples).

> 2) what the characteristics are of the firms when you take just one
> variable
> and its descriptive statistics. For instance: you take alliance
> experience
> as a variable.

I assume that you realize that this depends on the type of variable
or level of measurement of the variable you are using.

Let's assume that "alliance experience" is a 5 level ordinal scale
variable.
What you seem to be asking for is something like a brealdown table for
each each level of "alliance experience" where all other atributes have
descriptive statistics (something that the SPSS "examine" or "explore"
or whatever the hell they are calling it these days) so one can compare
how a background variable changes across levels of "alliance
experience".
Some people are okay with this because they want the numerical
specificity
provided but this may be hard to follow in a table or tables.

If "alliance experience" is a 5 level ordinal scale variable, then
barcharts
can be used where, say, the mean or median of a background variable
is provided for each level.  One could use multiple barcharts to
represent
multiple background variables or there may be a way to have a
multivariate
graphic that shows multiple background variables for each level of
"alliance experience".

If "alliance experience" is a continuous variable, 2-D scatterplots with
other
variables can be used or there may be a multivariate analog that can be
used -- perhaps someone knows of such a graphic?

> The mean alliance experience is 4 (4 prior alliances). Then
> I'd like to know: looking at all these that on average had 4 prior
> alliances, how much do these firms spend on research and development
> (R&D)?
> It would be ideal to see a sort of visualization showing, for
> instance, that
> around the average of 4 alliances, firms were spending rather much on
> R&D,
> while firms with many prior alliances, were not spending much on R&D.

Knowing nothing about how many variables you have, whether you really a
dealing with a population or a sample, or why you may choose to look at
some background variables and not others, and so on, it is difficult to
provide good suggestions (others wiser than I may know better).

Again, if you really have a population, you are suggesting doing a
lot of different descriptive analysies (both numeric and graphic)
which will depend upon the nature of the variables being used
as well as whether you should use the raw values or some recoded
form (e.g., imagine "alliance experience" is actually a continuous
5 point scale, you may have to recod these values into 5 categories
"0.0-0.50", "0.5-1.5", "1.5-2.5". "2.5-3.5", "3.5-4.5", and "4.5-5.0";
you would then look at mean/median R&D spending or whatever
statistics you want for each level)..

I could be mistaken but I think you need to provide more info about your
dataset and what you think you're doing.  Then again, others with more
knowledge (as well as greater ESP ability) may be able to provide
more useful information.

> Thanks again, hope you can help!

For what it's worth.

-Mike Palij
New York University
[hidden email]

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