I used to think the same. Zotero is the only reference manager I've used that's not a pain in the arse - that is, it's much easier to use it than it is not to use it.
Zotero is the best piece of software I've come across in a long time, and the database schema is particularly nice. I always thought that Thomposon were fools. Now on one side they're having their lunch eaten by google scholar, and on the other side by a variety of free and/or open source bibliographic managers.
For any Thomposon execs reading - if you don't stop regarding the users data as your property and start opening up instead, your decline will be much faster than similar proprietary software companies.
This seems as good a place as any to let interested people know about the 2007 Catalyst Advent Calendar. 24 days of how to work with catalyst. This year things are getting pretty sophisticated, with the first two entries dealing with Catalyst/YUI integration, including how to build a CRUD app with catalyst and YUI.
The difference is that PHP will eventually decline due to drowning in its own cruft. Perl doesn't have the same architectural problems, and so is well poised for a resurgance. Besides there's a ton of legacy and not-so-legacy code out there, it's just that people tend to keep quiet about it for some reason (maybe commercial in confidence?)
The outcome may be the same, but the process by which the outcome was reached may differ, so my findings stand, they're just a bit more subtle than I implied originally.
Ok I had a look at TFA and I decided they were running the wrong test. What we're actually looking at is whether the corruption index is dependent on the vote. I fired up R and ran a quick kruskal-wallis test (1 way non parametric analysis of variance):
> ct kruskal.test ( ct$CPI ~ ct$Vote)
Kruskal-Wallis rank sum test
data: ct$CPI by ct$Vote
Kruskal-Wallis chi-squared = 21.531, df = 3, p-value = 8.166e-05
So there is a significant difference between voting behaviour by corruption index. However this doesn't tell us the size and direction of the difference, so I ran a quick notched boxplot of vote against CPI:
plot(ct$CPI ~ ct$Vote, notch=T)
The graph is here. If you look at where the notches in the boxplots don't overlap, that will tell you where the significant difference lies. It turns out that the median corruption index score for the yes votes is significantly lower than for all the other groups. This indicates that there is an association between corruption perception and voting yes (i.e. yes voters are seen as more corrupt). This appears to be a very strong significant difference and not the marginally significant difference as in TFA at all.
Actually Fisher's exact test is used in exactly the same situation as chi squared except when samples are small. To be honest what I'd do here is a spearmans rank correlation (for non parametric data) and have a look from there. One of the adavantages of this is that correlation coefficients are basically meaningless from the point of view of statistical significance (being exclusively proportional to sample size) so you can basically handwavingly say that the correlation is small, medium or large and positive or negative without having to draw dubious conclusions about marginal statistical significance.
Slashdot Mirror
You must have one of these mutations that leaves your hands with quite a lot of fingers. 6700 CPAN authors at last count.
I used to think the same. Zotero is the only reference manager I've used that's not a pain in the arse - that is, it's much easier to use it than it is not to use it.
Zotero is the best piece of software I've come across in a long time, and the database schema is particularly nice. I always thought that Thomposon were fools. Now on one side they're having their lunch eaten by google scholar, and on the other side by a variety of free and/or open source bibliographic managers. For any Thomposon execs reading - if you don't stop regarding the users data as your property and start opening up instead, your decline will be much faster than similar proprietary software companies.
This seems as good a place as any to let interested people know about the 2007 Catalyst Advent Calendar. 24 days of how to work with catalyst. This year things are getting pretty sophisticated, with the first two entries dealing with Catalyst/YUI integration, including how to build a CRUD app with catalyst and YUI.
nethack > pong > everything else
The difference is that PHP will eventually decline due to drowning in its own cruft. Perl doesn't have the same architectural problems, and so is well poised for a resurgance. Besides there's a ton of legacy and not-so-legacy code out there, it's just that people tend to keep quiet about it for some reason (maybe commercial in confidence?)
The outcome may be the same, but the process by which the outcome was reached may differ, so my findings stand, they're just a bit more subtle than I implied originally.
Ok I had a look at TFA and I decided they were running the wrong test. What we're actually looking at is whether the corruption index is dependent on the vote. I fired up R and ran a quick kruskal-wallis test (1 way non parametric analysis of variance): > ct kruskal.test ( ct$CPI ~ ct$Vote) Kruskal-Wallis rank sum test data: ct$CPI by ct$Vote Kruskal-Wallis chi-squared = 21.531, df = 3, p-value = 8.166e-05 So there is a significant difference between voting behaviour by corruption index. However this doesn't tell us the size and direction of the difference, so I ran a quick notched boxplot of vote against CPI: plot(ct$CPI ~ ct$Vote, notch=T) The graph is here. If you look at where the notches in the boxplots don't overlap, that will tell you where the significant difference lies. It turns out that the median corruption index score for the yes votes is significantly lower than for all the other groups. This indicates that there is an association between corruption perception and voting yes (i.e. yes voters are seen as more corrupt). This appears to be a very strong significant difference and not the marginally significant difference as in TFA at all.
Actually Fisher's exact test is used in exactly the same situation as chi squared except when samples are small. To be honest what I'd do here is a spearmans rank correlation (for non parametric data) and have a look from there. One of the adavantages of this is that correlation coefficients are basically meaningless from the point of view of statistical significance (being exclusively proportional to sample size) so you can basically handwavingly say that the correlation is small, medium or large and positive or negative without having to draw dubious conclusions about marginal statistical significance.