jif-vs-citations.py

%matplotlib qt
from scipy.stats import pearsonr,spearmanr
from scipy import stats
from matplotlib import pyplot as plt
import pandas as pd
import numpy as np
import seaborn as sns

data = pd.read_excel('./062109-1.xlsx', sheet_name=2)
data = data[data.columns[:12]]
data.set_index(data.columns[0], inplace=True)
data.fillna(0, inplace=True)

citations = data.reset_index()['Citations'].replace('100+', 100).astype(int)
jif_est = {}
for c in data.columns:
    jif_est[c] = (data[c].values*citations).sum()/data[c].sum()

cumdata = data.cumsum(axis=0)

def random_select(col):
    p = np.random.randint(cumdata[col].iloc[-1])
    return np.searchsorted(cumdata[col], p)


def pair_compare(col1, col2):
    c1 = random_select(col1)
    c2 = random_select(col2)
    return c1 < c2


def pair_compare_many(col1, col2, n):
    return sum(pair_compare(col1, col2) for _ in range(n)) / n

low = [c for c in data.columns if jif_est[c] < 5]
medium = [c for c in data.columns if 5 <= jif_est[c] < 15]
high = [c for c in data.columns if 15 <= jif_est[c]]

compares = []
for ix,c1 in enumerate(data.columns):
    for c2 in data.columns[ix+1:]:
        compares.append((c1, c2, pair_compare_many(c1, c2, 10_000)))
compares = pd.DataFrame(compares, columns=['Journal 1', 'Journal 2', 'P(J1->citations)'])
print(compares)

def select_category(cat):
    if cat == 'low':
        return np.random.choice(low)
    elif cat == 'medium':
        return np.random.choice(medium)
    elif cat == 'high':
        return np.random.choice(high)
    else:
        raise ValueError(f'Unknown category {cat}')

def compare_category(cat1, cat2):
    return pair_compare_many(select_category(cat1), select_category(cat2), 10_000)


cat_comps = []
for c1 in ['low', 'medium', 'high']:
    for c2 in ['low', 'medium', 'high']:
        cat_comps.append((c1, c2, compare_category(c1, c2)))

cat_comps = pd.DataFrame(cat_comps, columns=['Category 1', 'Category 2', 'P(JIF->citations)'])
print(cat_comps)

jif_prob = []
for ix, c1 in enumerate(data.columns):
    for c2 in data.columns[ix+1:]:
        jif_prob.append((jif_est[c2] - jif_est[c1], pair_compare_many(c1, c2, 10_000)))

jif_prob = np.array(jif_prob)
np.corrcoef(jif_prob[:,0], jif_prob[:,1])

fig, ax = plt.subplots()
ax.scatter(jif_prob[:,0], jif_prob[:,1])
ax.set_xlabel('JIF difference (estimated)')
ax.set_ylabel('P(random paper in higher JIF journal has more citations)')
sns.despine(fig, trim=True)
fig.tight_layout()
fig.savefig('jif_prob.png')
fig.savefig('jif_prob.svg')

sanity = []
for c in data.columns:
    simul = [random_select(c) for _ in range(10_000)]
    sanity.append((c, jif_est[c], np.mean(simul), np.median(simul)))
sanity = pd.DataFrame(sanity, columns=['Journal', 'JIF estimate', 'Random citation count', 'Median citation count'])
print(sanity)
stats.spearmanr(sanity['Median citation count'], sanity['Random citation count'])


jif_cites = []
for c in data.columns:
    for _ in range(100):
        jif_cites.append((c, jif_est[c], random_select(c)))

jif_cites = pd.DataFrame(jif_cites, columns=['Journal', 'JIF estimate', 'Citations'])
spearmanr(jif_cites['JIF estimate'], jif_cites['Citations'])
fig, ax = plt.subplots()
ax.clear()
ax.scatter(jif_cites['JIF estimate'], jif_cites['Citations'], alpha=0.1)