Background & Summary

The replicability of findings from scientific research has garnered enormous popular and academic attention in recent years1–3. Results of replication initiatives attempting to reproduce previously published findings reveal that the majority of independent studies do not produce the same significant effects as the original investigation1–5.

There are many reasons why a scientific study may fail to replicate besides the original finding representing a false positive due to publication bias, questionable research practices, or error. These include meaningful population differences between the original and replication samples (e.g., cultural, subcultural, and demographic variability), overly optimistic estimates of study power based on initially published results, study materials that were carefully pre-tested in the original population but are not as well suited to the replication sample, a lack of replicator expertise, and errors in how the replication was carried out. Nonetheless, the low reproducibility rate has contributed to a crisis of confidence in science, in which the truth value of even many well-established findings has suddenly been called into question6.

The present line of research introduces a collaborative approach to increasing the robustness and reliability of scientific research, in which findings are replicated in independent laboratories before, rather than after, they are published7,8. In the Pre-Publication Independent Replication (PPIR) approach, original authors volunteer their own findings and select expert replication labs with subject populations they expect to show the effect. PPIR increases the informational value of unsuccessful replications, since common alternative explanations for failures to replicate such as a lack of replicator expertise and theoretically anticipated population differences are addressed. Sample sizes are also much larger than is common in the field, and the analysis plan is pre-registered9, allowing for more accurate effect size estimates and identification of unexpected population differences. An effect has been overestimated and is quite possibly a false positive if it consistently fails to replicate in PPIRs. Pre-publication independent replication also has the benefit of ensuring published findings are reliable before they are widely disseminated, rather than only checking after-the-fact.

In this first crowdsourced Pre-Publication Independent Replication (PPIR) initiative, 25 laboratories attempted to replicate 10 unpublished moral judgment findings in the research ‘pipeline’ of the last author and his collaborators (see Table 1). The original authors selected replication laboratories with directly relevant expertise (e.g., moral judgment researchers), and access to subject populations theoretically expected to show the effect. A pre-set list of replication criteria were applied10: whether the original and replication effect were in the same direction, whether the replication effect was statistically significant, whether the effect size was significant meta-analyzing the original and replication studies, whether the original effect size was within the confidence interval of the replication effect size, and finally the small telescopes criterion (a replication effect size large enough to be reliably captured by the original study11). Of the 10 original findings, 6 replicated according to all criteria, two studies failed to replicate entirely, one study replicated but with a smaller effect size than the original study, and one study replicated in United States samples but not outside the United States (see ref. 7 for a full empirical report).

Table 1 Overview of Replications.
Full size table

Unique among the replication initiatives thus far, the pipeline project corpus includes the data from not only the replications but also all of the original studies targeted for replication. This creates a unique opportunity for future analysts to better understand reproducibility and irreproducibility in science, since the data from the original studies can be reanalyzed to better understand why a particular effect did or did not prove reliable. The dataset is complemented by both socioeconomic and demographic information on the research participants, and contains data from 6 countries (the United States, Canada, the Netherlands, France, Germany, and China) and replications in 4 languages (English, French, German, and Chinese). The Pre-Publication Independent Replication Project dataset is publicly available on the Open Science Framework (Data Citation 1) and is accompanied by SPSS syntax which can be used to reproduce the analyses. This array of data will serve as a resource for researchers interested in research reproducibility, statistics, population differences, culturally-moderated phenomena, meta-science, moral judgments, and the complexities of replicating studies. For example, the data can be re-analyzed using meta-regression techniques in order to better understand if certain study characteristics or demographics moderate effect sizes. A re-analyst could also try out different analytic techniques and see how robust certain effects are to different specifications.

Methods

Participants

The Pre-Publication Independent Replication Project corpus includes three datasets. The first dataset (PPIR 1.sav: Data Citation 1) contains data from 3 original studies and their replications, a second dataset (PPIR 2.sav: Data Citation 1) contains data from 3 original studies and their replications, and a third dataset (PPIR 3.sav: Data Citation 1) contains data from 4 original studies and their replications. In total data were collected from 11,805 participants. The first SPSS file (PPIR 1.sav: Data Citation 1) contains data from 3,944 participants (including 514 from the original studies), while the second SPSS file (PPIR 2.sav: Data Citation 1) contains data from 3,919 participants (including 351 from original studies) and the final SPSS file (PPIR 3.sav: Data Citation 1) contains data from 3,829 participants (including 582 from original studies). An additional replication dataset collected in France contained 113 participants. No participants were removed from either the original or replication studies. All participants agreed to the informed consent form and the studies were in accordance with ethics regulations of the respective universities.

Testing procedure

The data were collected using both online and paper-pencil surveys from the respective laboratories and participants. The replications used the same materials and measurements as in the original studies, with the exception that the materials were translated into multiple languages. In the online version of the replications, Qualtrics was used to collect the data. This online platform allowed us to randomize the order by which the studies were presented. In order to prevent participant fatigue, studies were administered in one of three batches, each batch contained three to four studies, and study order was counterbalanced between subjects. Once subjects agreed to participate in the study, they read vignettes (see below for an example of the vignette from the Cold-Hearted Prosociality study) and completed survey questions assessing their reactions. Thereafter, the participants were thanked for their participation and debriefed.

Karen works as an assistant in a medical center that does cancer research. The laboratory develops drugs that improve survival rates for people stricken with breast cancer. As part of Karen’s job, she places mice in a special cage, and then exposes them to radiation in order to give them tumors. Once the mice develop tumors, it is Karen’s job to give them injections of experimental cancer drugs.

Lisa works as an assistant at a store for expensive pets. The store sells pet gerbils to wealthy individuals and families. As part of Lisa’s job, she places gerbils in a special bathtub, and then exposes them to a grooming shampoo in order to make sure they look nice for the customers. Once the gerbils are groomed, it is Lisa’s job to tie a bow on them.

Although the majority of the data were collected as described above, there were some exceptions. Specifically, as opposed to counterbalancing the order in which the study was presented, participants at Northwestern University were randomly allocated to a survey which either contained one longer study or three shorter studies that were presented in a fixed order. Participants at Yale University did not complete one study as the researchers felt that the participants may be offended by it. Also, there was a translation error in one study run at the INSEAD Paris laboratory which required that study to be re-run separately. Finally, study order for participants at HEC Paris was not counterbalanced but rather fixed. Table 2 includes an outline of the number of replications and conditions, a brief synopsis of study and instructions for creating variables. Detailed reports of each original study and the complete replication materials are available on the OSF in the Supplementary File 1 (00.Supplemental_Materials_Pipeline_Project_Final_10_24_2015.pdf: Data Citation 1). Supplementary File 2 outlines all the names and measurement details used in the study (PPIR _Codebook.xlsx: Data Citation 1).

Table 2 Technical Validation and Study Synopsis.
Full size table

Data Records

All data records listed in this section are available from the Open Science Framework (Data Citation 1) and can be downloaded without an OSF account. The datasets were anonymized to remove any information that could identify the participant responses, such as identification numbers from Amazon’s Mechanical Turk. The analysis was conducted with SPSS version 20 and detailed SPSS syntax (including comments) are provided to help with data analysis. In total there are 3 datasets and 11 syntax files available. These datasets are also accompanied by a codebook which describes the variables, the coding transformations necessary to replicate the analyses, and a synopsis of the respective studies.

First dataset

Location: (PPIR 1.sav: Data Citation 1)

File format: SPSS Statistic Data Document file (.sav)

This file contains basic demographic information and responses to the items measured in the Moral Inversion study (SPSS Syntax files/PPIR 1–2 moral inversion.sps: Data Citation 1), Intuitive Economics study (SPSS Syntax files/PPIR 1–4 intuitive economics.sps: Data Citation 1), and Burn in Hell study (SPSS Syntax files/PPIR 1–7 burn in hell.sps: Data Citation 1).

Second dataset

Location: (PPIR 2.sav: Data Citation 1)

File format: SPSS Statistic Data Document file (.sav)

This file contains basic demographic information and responses to the items measured in the Presumption of Guilt study (SPSS Syntax files/PPIR 2—1 presumption of guilt.sps: Data Citation 1), The Moral Cliff study (SPSS Syntax files/PPIR 2–3 moral cliff.sps: Data Citation 1), and Bad Tipper study (SPSS Syntax files/PPIR 2–9 bad tipper.sps: Data Citation 1).

Third dataset

Location: (PPIR 3.sav: Data Citation 1)

File format: SPSS Statistic Data Document file (.sav)

This file contains basic demographic information and responses to the items measured in the Higher Standard Effect study (SPSS Syntax files/PPIR 3–5 higher standard—Charity.sps: Data Citation 1; SPSS Syntax files/PPIR 3–5 higher standard—Company.sps: Data Citation 1), Cold Hearted Prosociality study (SPSS Syntax files/PPIR 3–6 cold-hearted.sps: Data Citation 1), Bigot-Misanthrope study (SPSS Syntax files/PPIR 3–8 bigot misanthrope.sps: Data Citation 1), and Belief-Act Inconsistency study (SPSS Syntax files/PPIR 3–10 belief-act inconsistency.sps: Data Citation 1).

Codebook

Location: (Data descriptor—Codebook/PPIR _Codebook.xlsx: Data Citation 1)

File format: Microsoft Excel Worksheet (.xlsx)

Introduction to PPIR project, outline of transformations, descriptions and labels for variables from three datasets.

Technical Validation

The studies include an array of original measurements which must be calculated to test the concepts of interest. These measures range from a single item to aggregated measures with multiple items, some of which must be reverse coded. (See Fig. 1, for an example of the items measuring candidate evaluations in the Higher Standards study; note that item 5 must be reverse coded prior to averaging the items into a composite). Instructions for how to create the study variables, the relevant conditions, and a synopsis of what concepts the variables measure, can be found in Table 2.

Figure 1: Example of the items measuring a typical moral judgement effect - in this instance candidate evaluations in the Higher Standards study.
figure 1

Figure 1 outlines a typical questionnaire that was administered to the subjects to assess their attitudes and beliefs toward the characters depicted in the vignettes. The subjects were required to write next to the statement the number that best indicated how much they believed the statement was representative of Lisa’s or Karen’s characteristics.

Full size image

Figure 1 outlines a typical questionnaire that was administered to the subjects to assess their attitudes and beliefs toward the characters depicted in the vignettes. The subjects were required to write next to the statement the number that best indicated how much they believed the statement was representative of Lisa’s or Karen’s characteristics.

Additional Information

How to cite this article: Tierney, W. et al. Data from a pre-publication independent replication initiative examining ten moral judgement effects. Sci. Data 3:160082 doi: 10.1038/sdata.2016.82 (2016).