Replications of Published Findings: PCS2
We have conducted replication analyses of four of the publications that derived from PCS2 data:
- Cohen, S., Doyle, W. J., Turner, R. B., Alper, C. M., & Skoner, D. P. (2003)*. Emotional style and susceptibility to the common cold. Psychosomatic Medicine, 65, 652-657. (variable list 1)
- Cohen, S., Doyle, W. J., Turner, R. B., Alper, C. M., & Skoner, D. P. (2003)*. Sociability and susceptibility to the common cold. Psychological Science,14, 389-395. (variable list 2)
- Cohen, S., Doyle, W. J., Turner, R., Alper, C. & Skoner, D. P. (2004). Childhood socioeconomic status and host resistance to infectious illness in adulthood. Psychosomatic Medicine, 66, 553-558. (variable list 3)
- Doyle, W. J., Gentile, D. A., & Cohen, S. (2006)*. Emotional style, nasal cytokines and illness expression after experimental rhinovirus exposure. Brain, Behavior and Immunity, 20, 175-181. (variable list 4)
Inconsistencies between results from 3 of these publications (indicated with an asterisk) and the PCS2 data replications are described in detail below. For additional information on the data set variables that were used in each analysis, click on the links indicated above.
Inconsistencies Between Results Published in Cohen et al (2003, Psychosom Med) and PCS2 Data Set Replication
Total adjusted symptom scores. For the analyses reported by Cohen et al (2003), missing symptom score data were handled by replacing missing value(s) with the average of the values reported for the surrounding days. For example, if a participant was missing a total symptom score for Day 2, that missing value would be replaced with the average of the values reported on Day 1 and Day 3. Thus, total adjusted scores were computed by summing all adjusted values (including those that were imputed) from Day 1 through Day 5. For the PCS2 Data Set, missing data were left as missing, with total adjusted scores being computed by taking the mean of all non-missing adjusted symptom scores, and then multiplying by 5.
Subjective colds. Though not explicitly reported by Cohen et al (2003), the number of participants who met criteria for a subjective cold (n = 103) can be determined from the data presented in Table 1 (p. 655). Using the data contained in the PCS2 Data Set, the frequency of subjective colds is n = 104. The discrepancy results from the slight difference in total adjusted symptom scores described above.
Inconsistencies Between Results Published in Cohen et al (2003, Psychol Sci) and PCS2 Data Set Replication
Total adjusted symptom scores. For the analyses reported by Cohen et al (2003), missing symptom score data were handled by replacing missing value(s) with the average of the values reported for the surrounding days. For example, if a participant was missing a total symptom score for Day 2, that missing value would be replaced with the average of the values reported on Day 1 and Day 3. Thus, total adjusted scores were computed by summing all adjusted values (including those that were imputed) from Day 1 through Day 5. For the PCS2 Data Set, missing data were left as missing, with total adjusted scores being computed by taking the mean of all non-missing adjusted symptom scores, and then multiplying by 5.
Subjective colds. Because of the slight difference in total adjusted symptom scores described above, the number of participants meeting the subjective cold criterion differs between the archived PCS2 data set and the data set used for the analyses reported in the publication (n = 104 vs. n = 103, respectively).
Daily social interaction variables from interviews. Four of the interview variables that were used in the analyses reported by Cohen et al (2003) were computed for participants in trials 1 through 9, but not for trial 10.
- Average number of people interacted with per day
- Average pleasantness of interactions
- Percentage of interactions that involved moderate to severe conflict
- Number of interactions that involved moderate to severe conflict
Accordingly, the published correlations involving these variables reflect only the first 9 trials (see Table 4, p. 393). Correlations using data from all 10 trials do not differ appreciable from those using the reduced sample size.
An additional four social interaction variables differ from their analogs in the PCS2 Data Set due to an error that occurred when the variables originally were created.
- Percentage of interactions that were pleasant
- Number of interactions that were pleasant
- Percentage of interactions that were unpleasant
- Number of interactions that were unpleasant
Response options for the ratings of social interaction pleasantness ranged from 1 (unpleasant) to 7 (pleasant). For purposes of analysis, pleasant interactions were defined as those with ratings 5, 6, or 7; unpleasant interactions were defined as those with ratings 1, 2, or 3 (see p. 390, Social interactions). When the pleasant and unpleasant variables actually were created, however, the syntax established cut points based on a range of 0 to 6 rather than 1 to 7 (pleasant criteria = 4, 5, or 6; unpleasant criteria = 0, 1, or 2). Accordingly, interactions with ratings of 3 were excluded from the unpleasant interactions count. Likewise, interactions with ratings of 7 were excluded from the pleasant interactions count, while interactions with ratings of 4 were included.
Inconsistencies Between Results Published in Doyle et al (2006, BBI) and PCS2 Data Set Replication
Post-challenge cytokine variables. Information regarding the specific formula used to compute cytokine area-under-the-curve (AUC) values was not available. Using the average adjusted post-challenge cytokine scores in the analyses provides a close approximation to the published findings. In most cases, discrepancies lean in favor of the analyses conducted using the PCS2 data set resulting in stronger associations.
Positive emotional style (PES). The description of how PES was calculated is incorrect. The variable used in the analyses reported by Doyle et al (2006) was computed as the mean of the average positive affect scores derived from three separate administrations of a self-report trait affect questionnaire. However, results are essentially identical when the PES variable that was used in Cohen et al (2003, Psychosom Med) is substituted into the models.