Brian Wansink is quite possibly the worst researcher you’ve (not) heard of. Due to the attention to his works as of lately, there are now 42 reports on his publications. Here we use these for citation analysis to see if science rewards bad practices.
options(digits=2)
library(pacman)
p_load(kirkegaard, readr, dplyr, rms)#d = read.table("clipboard", sep = "\t", header=T, quote = "", stringsAsFactors = F)
#write_rds(d, "data/data.rds")
d = read_rds("data/data.rds")
#recoder func
recode_this = function(x) {
x %>%
str_trim %>%
plyr::mapvalues(from = c("", "+", "++"), to = c(0, 1, 2), warn_missing = F) %>%
as.integer
}
#recode
d = d %>% dplyr::mutate(
self_plagiarism = self_plagiarism %>% recode_this,
data_duplication = data_duplication %>% recode_this,
data_issues = data_issues %>% recode_this,
statistical_issues = statistical_issues %>% recode_this,
year = str_match(publication, pattern = "\\d{4}") %>% as.vector %>% as.integer,
age = (2017 - year)
)
#total issues
issue_types = c("self_plagiarism", "data_duplication", "data_issues", "statistical_issues")
d$total_issues = rowSums(d[issue_types])#cors
d[map_lgl(d, ~is.numeric(.))] %>% wtd.cors## citations self_plagiarism data_duplication data_issues
## citations 1.000 -0.074 -0.19 -0.135
## self_plagiarism -0.074 1.000 0.19 -0.562
## data_duplication -0.191 0.190 1.00 0.223
## data_issues -0.135 -0.562 0.22 1.000
## statistical_issues -0.210 -0.565 0.20 0.829
## year -0.247 -0.179 -0.17 0.071
## age 0.247 0.179 0.17 -0.071
## total_issues -0.280 -0.055 0.71 0.746
## statistical_issues year age total_issues
## citations -0.21 -0.247 0.247 -0.280
## self_plagiarism -0.56 -0.179 0.179 -0.055
## data_duplication 0.20 -0.172 0.172 0.711
## data_issues 0.83 0.071 -0.071 0.746
## statistical_issues 1.00 0.109 -0.109 0.739
## year 0.11 1.000 -1.000 -0.060
## age -0.11 -1.000 1.000 0.060
## total_issues 0.74 -0.060 0.060 1.000#age only
lm(citations ~ age, data = d) %>% summary##
## Call:
## lm(formula = citations ~ age, data = d)
##
## Residuals:
## Min 1Q Median 3Q Max
## -125.9 -71.7 -38.3 -6.1 774.6
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 31.84 42.81 0.74 0.46
## age 6.50 4.04 1.61 0.12
##
## Residual standard error: 150 on 40 degrees of freedom
## Multiple R-squared: 0.0608, Adjusted R-squared: 0.0373
## F-statistic: 2.59 on 1 and 40 DF, p-value: 0.115#total data issues
lm(citations ~ age + total_issues, data = d) %>% summary##
## Call:
## lm(formula = citations ~ age + total_issues, data = d)
##
## Residuals:
## Min 1Q Median 3Q Max
## -171.8 -82.8 -31.2 30.4 726.1
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 97.11 52.69 1.84 0.073 .
## age 6.97 3.90 1.78 0.082 .
## total_issues -22.77 11.42 -1.99 0.053 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 145 on 39 degrees of freedom
## Multiple R-squared: 0.148, Adjusted R-squared: 0.104
## F-statistic: 3.38 on 2 and 39 DF, p-value: 0.0443By traditional NHST we almost found evidence that science does not reward QRPs, the beta is negative and almost significant, so just maybe readers did notice some of the problems and decided not to cite in return? Or, more cynically, maybe Wansink is just avoiding citing his own poor work to not draw attention. One would have to separate the citations into self and other citations to tell.
Of course, the sample size is too small; we could not even find a reliable effect of publication age. But Wansink has a lot of papers, so we (that means not me) can expand the dataset by analyzing more of his studies.