#!/usr/bin/python

### Calculate and graph correlation between publication fee and
### eigenfactor for open-access journals.
###
### Stuart M. Shieber
### 10/12/2009

### The correlation coefficients are useful for exploring the issue of
### whether open-access journals operate as a vanity publishing
### industry (like trade-book publishing) in which standards are
### inversely correlated with author fees (that is, positively
### correlated with royalties, which are negative author fees) or as a
### quality publishing industry, in which standards are positively
### correlated with author fees.

### This program assumes that the journal data is stored in a
### comma-separated value file journal-data.csv.  It prints a small
### report on the number of journals used in the analysis and the
### correlation coefficient (Pearson's) between the publication fees
### and article influences.  It also generates a scatter plot with
### linear fit lines of the data written to vanitypress.png.

### For further discussion, see The Occasional Pamphlet, "Is
### open-access publishing a vanity publishing industry?"

import csv                      # i/o for csv files
import numpy                    # for statistics
import matplotlib               # for plotting
matplotlib.use('Agg')
import matplotlib.pyplot as plt

### Utilities

# isFloat: test if a string represents a floating point number
def isFloat(str):
    try:
        float(str)
        return True
    except:
        return False

### Currency conversion

# Currency conversion rates as of 1/13/2009 as provided by oanda.com

rates = { "CAD":	0.83523,
          "CHF":	0.89588,
          "EUR":	1.34011,
          "GBP":	1.49963,
          "USD":	1,
          "CNY":	0.14648  }

### Get journal data

reader = csv.DictReader(open('journal-data.csv', 'rU'))

charges = []
influences = []
chargesNZ = []
influencesNZ = []

for journal in reader:

    # Filter out some journals that aren't in the analysis.  We only
    # include journals that charge publication fees, that don't charge
    # submission fees, and that have an eigenfactor-based article
    # influence value.
    if (not isFloat(journal['Publication Charge Per Article']) or
        (isFloat(journal['Submission Charge']) and
         float(journal['Submission Charge']) > 0) or
        not isFloat(journal['Article Influence'])       ):
        continue

    # Find the proper exchange rate
    if (journal['Currency'] == "" or
        journal['Currency'] == "NA"):
        #print 'No currency spec?', journal['Publication Charge Per Article'], journal['Title']
        rate = 1
    else:
        rate = rates[journal['Currency']]

    # Compute the publication fee in USD
    charge = float(journal['Publication Charge Per Article']) * rate

    # Add the data point
    charges.append(charge)
    influences.append(float(journal['Article Influence']))
    if (charge > 0):
        # separately for non-zero charges
        chargesNZ.append(charge)
        influencesNZ.append(float(journal['Article Influence']))
        
### Report correlation coefficients

print "Based on %d journals (%d non-zero fee)" % (len(charges), len(chargesNZ))
print "Correlation for all:       %5.3f" % numpy.corrcoef([charges, influences])[0][1]
print "Correlation for non-zero:  %5.3f" % numpy.corrcoef([chargesNZ, influencesNZ])[0][1]

### Generate scatter plot
plt.scatter(charges, influences)

# Add best-fit line for all and non-zeros
slope, intercept = numpy.polyfit(charges,influences, 1)
slopeNZ, interceptNZ = numpy.polyfit(chargesNZ,influencesNZ, 1)

plt.plot([0,2850], [intercept, slope*2850+intercept], 'g:')
plt.plot([0,2850], [interceptNZ, slopeNZ*2850+interceptNZ], 'b:')

# Add axes
plt.axis([-50, round(max(charges)/100+1)*100,
          -.25, round(max(influences))+.25])

# Label axes
plt.xlabel('Publication Charge (US$)', family='serif')
plt.ylabel('Article Influence', family='serif')

             
# Write out the figure
plt.savefig('vanitypress')