Fig 1.
Distributions of Importance of Olfaction Questionnaire (IO-Q) scores.
Raincloud plots [22] of the score distributions for total IO-Q (A) and subscales (B-E) for tests 1 and 2. The ‘cloud’ illustrates the data distribution, while below, observations are marked as jittered dots. On the dots, a boxplot is placed.
Fig 2.
Internal consistency and comparison between Tests 1 and 2.
A) Bar-charts showing (A) Cronbach’s alpha for the IO-Q (error bars represent 95% confidence intervals) for tests 1 and 2 along with the value reported by Croy et al. (2010). B-F) Mean scores for tests 1 and 2 for total IOQ and subscales including the means reported by Croy et al. (2011). Error bars in B-F represent the standard deviations from the mean.
Table 1.
Descriptive table of all scores.
Table 2.
Bayesian paired t-test.
Table 3.
Bayesian two-sample t-test.
Fig 3.
Correlation matrices for total IO-Q and subscale scores.
A and C: Internal correlation matrices for test 1 (A) and test 2 (C) using Bayesian statistics with a uniform non-directional prior. F: Correlation matrix for values reported by Croy et al. (2010) for comparison (they did not report the Agg subscale). B, D, and E: Matrices reporting log(BF10) for replicative correlational analysis. Arrows point toward the matrix used as a replicative correlation and away from the matrix used as a prior. A value in this matrix of 0 would correspond to equal odds of the null hypothesis versus the alternate hypothesis being true. Note they all results have log(BF10) > 4.6, corresponding to BF > 100, i.e., extreme evidence for replication.
Fig 4.
Summarizing table (top) and a scatterplot (bottom) showing Bayesian Pearson correlation between the total IO-Q in tests 1 and 2. Dots may overlap.
Table 4.
Bayesian correlation for subscales.
Fig 5.
Distribution of time between tests and a graphical posterior predictive check of the scaled difference between test score 1 and 2 versus time.
A) Histogram of the duration between tests 1 and 2. Bin width is 0.5 weeks. B) Using Bayesian statistics, a general linear model was fitted to the scaled difference between the total IO-Q score in tests 1 and 2 as a function of weeks between tests. If the general linear model is a good fit, then it should be able to produce data like currently observed data (y). Simulating an iterative run of 10,000 distributions (sample of those in light green, yrep), it is approximately predicting this data and is centered around zero. Kendall’s tau for the correlation between the scaled difference and time between tests was 0.078 (τ = 0.078, 95% CI = [-0.023;0.176], BF10 = 0.314), indicating a minimal influence of time.