TCATA Analysis

TCATA Analysis

Purpose

To produce descriptive summary statistics and statistical analyses particularly designed for TCATA data

Data Format

Datasets consist of samples tested by the panel, across one or more sessions, and samples can be repeated (replicates). Incomplete data is also possible. [See attached demo dataset]

Background

Temporal CATA data (TCATA) requires more specific analysis than is available in the temporal dominance curves analysis function previously available in EyeOpenR. Therefore this module provides analyses more appropriate for TCATA data, incorporating and extending analyses from the R package tempR plus additional unique analyses.

Options

The analyses can be performed attribute by attribute (comparing products) or product by product (comparing attributes)
  1. Alpha – specify the significance level you want to perform any statistical significance tests at (default=0.05)
  2. Smoothing – how much smoothing should be applied to the curve  - to use the unsmoothed curve (which will look like a series of steps as is based on count/proportion data) and so not apply any smoothing algorithm check ‘No’. This option is for the user to choose depending on their preference for appearance of the output. The “lower” the amount of smoothing chosen the closer to the original 0/1 data the curve will represent, however the appearance maybe hard to read.
  3. Result Display – Choose whether the output graphs use counts or proportions on the Y axis
  4. Onset Time – default set to T=30 seconds – From 0 to T seconds is the time period during  which the onset max % of checks per attribute will be derived
  5. Linger Time – default = T=120 second – this is used to calculate the area under the curve (AUC) after T seconds
  6. No. of decimals for values/p-values – as described, for the user to decide, useful defaults are given.

Results and Interpretation

When reading data in, it is particularly important for TCATA to check the information provided at  the ‘Quality Check’ step, where you can see the number of assessors by session, durations, variability measures etc in order to assess your data before proceeding with the analyses. Sometime issues in the analyses later on are purely due to the data being too sparse, too many zeros, no variability, missing data etc. so always check this section first to understand your data.
  1. Curves (Proportions) – Graph (one per attribute) showing the TCATA curve (count or proportion of assessors who checked the attribute on the Y-axis across time, for all the individual products). A table with the proportions at each time interval is shown underneath the curves and can be exported for further examination.
  2. Pairwise Differences (Proportions) – Graph showing the time period and count (or proportion) of checks that were statistically significantly different (at the significance level chosen) for attributes between the pairs of products being compared on each tab. If there are  no significant differences at the chosen specified significance level then the graph/tab will not be shown and a note is written explaining this in the information tab. No multiple comparison adjustments have been made for these tests. 
  3. KPIs – This summary table lists out 6 performance metrics, one row per product, attribute by attribute as follows so that comparisons can be made for each attribute across the products:
    1. Max (prop) – the maximum proportion of assessors who checked the attribute at any one time period
    2. Onset (prop) – Maximum % checked in first T seconds (default T=30 seconds unless specified in options)
    3. Onset (secs) – time until max perceived intensity (i.e. max proportion checked)
    4. Plateau** (area) – the area under the curve (AUC*) from 80-120% of maximum proportion i.e. max +/- 20%. This area is calculated by integrating all parts of the curve where the height is > 80% of maximum height. 
    5. Lingering (area) – AUC* after T seconds (where T is set in options but defaults to T=120 seconds)
    6. Attribute Factor (area) = whole AUC* from start of session to end
*Note on area calculations above: Area under curves (AUCs) are calculated by finding the timepoints where the curve height is as specified for the KPI (i.e. greater than or equal to 80% of the maximum height for the plateau). The integral is the sum of these curve heights multiplied by the time interval. Due to the nature of CATA data that the area may not be continuous as there may be separate regions where the 80% threshold is exceeded. i.e. there may be areas where no attributes may have been ticked for one of more time periods, (e.g. not checked at second 5-6, but were ticked at 2-4 and 7-9 seconds) which would create gaps in the integral calculation, so for this KPI the gaps have been ignored and a continuous curve has been assumed. The area calculations in the KPI table do not use any normalization process - they are computed directly from the average curves
  1. Max – These are the p-values for the statistical comparisons between products of the max proportion checked.  The proportions are compared using a Z-test. No multiple comparison adjustments have been made for these tests. Where counts for the 2 products are the same the pairwise test returns the value NA.
  2. Onset - Table of p-values for pairwise comparisons of onset time between each product pairing, attribute by attribute. The comparisons have been made using a 2-way ANOVA model (with interactions if replication is present) followed by pairwise testing. No multiple comparison adjustments have been made.
  3. Ring – This plot shows proportions of time an attribute of interest is checked for the products and compares them to each other in a ring chart, one per attribute. Factor proportions are calculated by considering the no. of times an attribute is checked for the product of interest (rcheck), compared to the number of times the attribute has been checked across ALL products(ncheck). The proportion is then rcheck/ncheck
  4. Linger – Table of p-values for pairwise comparisons of linger time between each product pairing, attribute by attribute. The comparisons have been made using a 2-way ANOVA model (with interactions if replication is present) followed by pairwise testing. No multiple comparison adjustments have been made.
  5. Plateau** Times – Table of start and stop times for each product plateau, attribute by attribute. 
  6. Plateau** ANOVA – ANOVA output for the Plateau AUC? Comparison of products AUC, attribute by attribute. The comparisons have been made using a 2-way ANOVA model (without interactions) followed by pairwise testing. The Tukey test has been used to make adjustments for multiple comparison testing.
  7. Plateau** P-values – Summary table showing the assessor and product P-values from the ANOVA comparison above
  8. Plateau** Means – Summary table showing the product AUC means for all attributes and their significance when compared in a pairwise fashion using the ANOVA model above. No multiple comparison adjustments have been made.
**Notes on normalisation of plateau curves:
Regarding the area and the related time limits for the plateau, t0 starts when the curve first crosses the 0.8 proportion and ends (t1) when it crosses for the last time the 0.8 proportion (considering all responses and regardless of possible drops in between below 0.8). The time span is clearly defined for each sample separately.  Within this time span per sample, individual proportions are calculated (how long within the defined time an assessor has the attribute selected). These individual proportions per sample are then corrected with the longest time span of all samples. These normalised curves are used for the plateau ANOVA analysis.

Technical Information

  1. R packages - tempR (https://cran.r-project.org/web/packages/tempR/tempR.pdf)
Additions and modifications have been coded by Qi Statistics to this package to extend and modify the functionality of tempR.

References

  1. Castura, J.C., Antúnez, L., Giménez, A., Ares, G. (2016). Temporal check-all-that-apply (TCATA): A novel temporal sensory method for characterizing products. Food Quality and Preference, 47, 79-90. doi: 10.1016/j.foodqual.2015.06.017 
  2. Meyners, M., Castura, J.C. (2018). The analysis of temporal check-all-that-apply (TCATA) data. Food Quality and Preference, 67, 67-76. doi: 10.1016/j.foodqual.2017.02.003

    • Related Articles

    • Penalty Analysis

      Purpose To provide a penalty analysis of a consumer data set, that is to investigate how liking or acceptability of product decreases when product attributes are not at the optimal intensity. Data Format Consumer.xlsx Note: for EyeOpenR to read your ...

    • Temporal Check All That Apply (TCATA)

      Introduction The Temporal Check All That Apply (TCATA) method builds upon the traditional CATA (Check All That Apply) question by introducing a dynamic element. It enables continuous selection and deselection of attributes based on their relevance to ...

    • Napping Analysis

      Purpose To provide an analysis of data collected using the napping methodology. Data Format Napping.xlsx For EyeOpenR to read your data the first five columns must include the following in the specified order: Assessor, Product, Session, Replica and ...

    • Multiple Factor Analysis (MFA)

      Purpose Performs a multiple factor analysis on a data set. This is not ‘Factor Analysis’ in the classic statistical sense, but rather a method for handling multiple groups of variables that are all measured on the same samples, and from this ...

    • Choice Base Conjoint Analysis

      Purpose This analysis method can be used to analyse data collected using the Choice Based Conjoint question type in EyeQuestion. Background Choice-Based Conjoint Analysis is a sophisticated market research technique used to decipher consumer ...