01-data-filtering.md

title	author	output	editor_options
Data Cleaning	Fernando Hoces de la Guardia	html_document	chunk_output_type console

Copy cleanned data sets (tidy df) from the SSRP_cleanining repo into this repo.

# tidy_dfs <- list.files(paste0(clean_path,"/processed/"))
# file.copy(paste0(clean_path,"/processed/",tidy_dfs), 
#           paste0("./processed/",tidy_dfs),
#           overwrite = TRUE)

First: basic clean up

• load data
• rename the title of a reproduction
• generate a variable that takes the same value for titles that sound similar
• ?generate unique identifier for each reproduction
• ?Identify the maximum number of claims and display items reproduced across all reproduction attempts

# dis_df <- read_csv("./processed/tidy_di_df.csv") %>%
#     mutate(
#         paper_title = ifelse(
#             test = # Fixing titles
#                 ( paper_title == "[Fernando] Railroads of the Raj: Estimating the Impact of Transportation Infrastructure"
#                 ),
#             yes = "Railroads of the Raj: Estimating the Impact of Transportation Infrastructure",
#             no = paper_title
#         ),
#         similar_title = phonetic(paper_title)
#     ) %>%
#     filter(!is.na(repro_score))

• Same but to claims-level data

# claims_df <- read_csv("./processed/tidy_claim_df.csv") %>%
#     mutate(paper_title= ifelse(
#         test = # Fix Fernando's title
#             (
#                 paper_title == "[Fernando] Railroads of the Raj: Estimating the Impact of Transportation Infrastructure"
#             ),
#             yes = "Railroads of the Raj: Estimating the Impact of Transportation Infrastructure",
#             no = paper_title
#         ),
#         similar_title = phonetic(paper_title)
#     ) # 98 reproductions (no abandoned)

• Same to abandoned papers

    # abandoned_df <- read_csv("./processed/tidy_abandoned_df.csv") %>%
    # mutate(paper_title= ifelse(
    #     test = # Fix Fernando's title
    #         (
    #             paper_title == "[Fernando] Railroads of the Raj: Estimating the Impact of Transportation Infrastructure"
    #         ),
    #         yes = "Railroads of the Raj: Estimating the Impact of Transportation Infrastructure",
    #         no = paper_title
    #     ),
    #     similar_title = phonetic(paper_title)
    # )  # 104 reproductions total

Make sure there's at least one scored Display Item

The rationale for this filter is that papers should have some sort of reproducibility score associated with them to be counted.

# has_score <- dis_df %>%
#     select(reproduction_id) %>%
#     unique()
# 
# claims_df <- claims_df %>%
#     inner_join(has_score,
#                by="reproduction_id")

Future Filters

This section explains how to add a filter in the future. Additional features should have a title (e.g. "Make sure there's at least one estimate associated with each claim") followed by a description of the reason we want this filter.

Code for the filter should reduce the number of observations in claims_df, dis_df, and/or abandoned_df, writing over the old df.

Temporary

FIXME: This section uses the clean data set used for working paper. Mimics all cleaning above. One change is that, particularly given the data from ssrp_analysis do not contain and use paper titles, but rather doi, codes are changed to use doi also.

# Load clean data
di_cl_df <- read.csv("data(temp)/di_cl_df_20240401.csv")
tidy_all_repro_df <- read.csv("data(temp)/tidy_all_repro_df_20240401.csv")
tidy_reproducer_df <- read.csv("data(temp)/tidy_reproducer_df_20240401.csv")


# Check if there is any missing (NA or "") in ID variables
di_cl_id_check <- di_cl_df %>% 
    filter((is.na(claim_N) | claim_N == "" | is.na(DI_N) | DI_N == "") & abandoned != 1) %>% 
    nrow()
stopifnot(di_cl_id_check == 0)

repro_paper_id_check <- tidy_all_repro_df %>% 
    filter(is.na(reproduction_id) | reproduction_id == "" | is.na(doi) | doi == "") %>% 
    nrow() 
stopifnot(repro_paper_id_check == 0)
    
# How many observations? (Using data from APril 1, 2024)
nrow(di_cl_df) #840 obs

[1] 840

nrow(tidy_all_repro_df) #260 obs

[1] 260

# How many papers, reproductions, claims, and DIs?

    # DI from completed repo (830)
    di_cl_df %>% 
        filter(abandoned == 0) %>%
        select(reproduction_id, claim_N, DI_N) %>%
        unique() %>% 
        nrow()

[1] 830

    # DI from abandoned repo (10)
    di_cl_df %>% 
        filter(abandoned == 1) %>%
        select(reproduction_id, claim_N, DI_N) %>%
        unique() %>% 
        nrow()

[1] 10

    # Claims from completed repo (500)
    di_cl_df %>% 
        filter(abandoned == 0) %>%
        select(reproduction_id, claim_N) %>%
        unique() %>% 
        nrow()

[1] 500

    # Claims from abandoned repo (10)
    di_cl_df %>% 
        filter(abandoned == 1) %>%
        select(reproduction_id, claim_N) %>%
        unique() %>% 
        nrow()    

[1] 10

    # Completed Reproductions (250)
    tidy_all_repro_df %>%
        filter(abandoned == 0) %>%
        select(reproduction_id) %>%
        unique() %>% 
        nrow()    

[1] 250

    # Abandoned Reproductions (10)
    tidy_all_repro_df %>%
        filter(abandoned == 1) %>%
        select(reproduction_id) %>%
        unique() %>% 
        nrow()       

[1] 10

    # Papers from completed reproductions (178)
    tidy_all_repro_df %>%
        filter(abandoned == 0) %>%
        select(doi) %>%
        unique() %>% 
        nrow()    

[1] 178

    # Papers from abandoned reproductions (8)
    tidy_all_repro_df %>%
        filter(abandoned == 1) %>%
        select(doi) %>%
        unique() %>% 
        nrow()     

[1] 8

    # Papers from all reproductions (186) - This doesn't need to necessarily be additive of above two.
    tidy_all_repro_df %>%
        select(doi) %>%
        unique() %>% 
        nrow()    

[1] 186