ANOVA with patents as DV
Locked
 
|
Dear all,
This might be a stupid question but I'm quite new to this. I want to run a ANOVA with one categorical independent variable (investment stage; 4 levels: Seed, Early, Expansion and Later) and my DV is number of patents. However, the way I understood using an ANOVA your DV should be either continuous, ordinal or dichotomous. My DV is # of patents, this is discrete, right? Can you still do a anova with a discrete DV like this? Thank you very much in advance. Kind regards, Maud Oostenbrink Maastricht University ===================== 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 |
|
|
Hi Maud,
A one way ANOVA seem the correct approach here (read up some of the many tutorials available online to become more familiar with this test), if indeed it is your intention to test the equality of the average number patents between the four different groups of the independent. Assuming the data collected for patents is the actual number of patents represented as a scale/continuous variable rather than banded into discreet bands i.e. 1-5, 6-10 ect ect. Hope this helps, Jignesh On 19 June 2013 12:08, Maud Oostenbrink <[hidden email]> wrote: Dear all, |
Administrator
|
In reply to this post by maudoostenbrink
For count DVs, it is common to use Poisson regression, or perhaps negative binomial regression if the counts are over-dispersed. There are also variations on these models that allow for having an overabundance of zeros (search on 'zero-inflated'). These models are run via GENLIN (Analyze > Generalized Linear Models > Generalized Linear Models).
But depending on your intended audience, those types of models may be difficult to digest. If it is reasonable and fair (i.e., not misleading) to use means and SDs descriptively, ANOVA will give you a fairly decent model too. Remember what George Box said about models -- all of them are wrong, but some are useful. HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
In reply to this post by maudoostenbrink
You do need to improve your understanding of the terminology.
Your IV, investment stage, is a set of categories that certainly do appear to be "ordinal" even though you may not want to use that fact in your analysis. From Seed to Later certainly does seem like a definite ordering in time. The usual classification includes "categorical" (or "nominal") which is unordered; "ordinal" which is ordered by not interval, where the rank-transformation allows tests that potentially are more efficient than ignoring the inherent order; "(equal) interval" which allows ANOVA; and "ratio" which has a natural zero, and mainly suggests the potential for natural power transformations of a scale that is ordinal into one that has "equal intervals" in respect to the dimension that is being modeled. ("Ratio" is a term that is seldom used usefully, and what I've suggested here is not what you will usually see.) Your DV of "number of patents" is apparently a count, so it has more potential uses than "discrete" = "nominal" = "unordered categories." What ANOVA requires for valid testing is a criterion IV that is reasonably close to "equal interval" so that the size of errors-in-prediction are not different for low versus middle versus high scores. In practice, this is judged by whether the "equal intervals" on the scale, from 0, 1, ... , 99, 100, ... are considered (a) equally predictable, and (b) of equal importance. Is the difference between 0 and 1 of similar "magnitude" as the difference between 99 and 100? A dichotomy is automatically "equal interval" since there is only one interval, so, Yes, it can be an IV in ANOVA. Contrary to your statement, an Ordinal variable is not automatically valid in ANOVA, as IV or as DV. It is labeled "ordinal" instead of "interval" *because* there is doubt about whether the intervals are equal. Being ordinal, it has potential to be transformed to ranks or by square root or log or what-have-you to become a variable that can be analyzed by ANOVA. - For samples that are not too tiny, the classical rank-order tests are essentially equivalent to ANOVA conducted on a rank-transformed version of the scores. Hope this helps. -- Rich Ulrich > Date: Wed, 19 Jun 2013 13:08:03 +0200 > From: [hidden email] > Subject: ANOVA with patents as DV > To: [hidden email] > > Dear all, > > This might be a stupid question but I'm quite new to this. I want to run a ANOVA with one categorical independent variable (investment stage; 4 levels: Seed, Early, Expansion and Later) and my DV is number of patents. However, the way I understood using an ANOVA your DV should be either continuous, ordinal or dichotomous. My DV is # of patents, this is discrete, right? Can you still do a anova with a discrete DV like this? > > Thank you very much in advance. > ... |
|
|
In reply to this post by Bruce Weaver
Zero-inflated count models, AFAIA, are not currently available in GENLIN.
Speaking of which, I continue to believe that offering a nonlinear mixed procedure in SPSS would be a game-changer. Ryan Sent from my iPhone On Jun 19, 2013, at 7:41 AM, Bruce Weaver <[hidden email]> wrote: > For count DVs, it is common to use Poisson regression, or perhaps negative > binomial regression if the counts are over-dispersed. There are also > variations on these models that allow for having an overabundance of zeros > (search on 'zero-inflated'). These models are run via GENLIN (Analyze > > Generalized Linear Models > Generalized Linear Models). > > But depending on your intended audience, those types of models may be > difficult to digest. If it is reasonable and fair (i.e., not misleading) to > use means and SDs descriptively, ANOVA will give you a fairly decent model > too. Remember what George Box said about models -- all of them are wrong, > but some are useful. > > HTH. > > > > maudoostenbrink wrote >> Dear all, >> >> This might be a stupid question but I'm quite new to this. I want to run a >> ANOVA with one categorical independent variable (investment stage; 4 >> levels: Seed, Early, Expansion and Later) and my DV is number of patents. >> However, the way I understood using an ANOVA your DV should be either >> continuous, ordinal or dichotomous. My DV is # of patents, this is >> discrete, right? Can you still do a anova with a discrete DV like this? >> >> Thank you very much in advance. >> >> Kind regards, >> >> Maud Oostenbrink >> Maastricht University >> >> ===================== >> To manage your subscription to SPSSX-L, send a message to > >> LISTSERV@.UGA > >> (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 > [hidden email] > http://sites.google.com/a/lakeheadu.ca/bweaver/ > > "When all else fails, RTFM." > > NOTE: My Hotmail account is not monitored regularly. > To send me an e-mail, please use the address shown above. > > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/ANOVA-with-patents-as-DV-tp5720812p5720813.html > Sent from the SPSSX Discussion mailing list archive at Nabble.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 ===================== 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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |