Experiment 2:

3.2.1. Demographics and Clinical Characteristics

Table 4 presents group comparisons of demographics and clinical characteristics.

The results showed main effects of age (F(3, 103) = 12.83, p < .001), education (F(3,

103) = 3.19, p = .027), IQ (F(3, 103) = 4.64, p = .004), and MMSE score (F(3, 103) =

10.41, p < .001) across four groups. Post hoc pairwise-comparison analyses using

Scheffe’s method indicated that the age of the HC group was younger than that of the

MCI and AD groups, and the age of SCD group was younger than that of AD group.

The education level in MCI group was significantly lower than that in the SCD.

Individuals in the HC and SCD groups showed higher IQ scores than did individuals

in AD group, whereas MCI group did not differ significantly from other groups. HC,

SCD, and MCI groups had higher scores on the MMSE than AD groups. No

differences in terms of other demographics, clinical characteristics, or

neuropsychological performance were found between HC and SCD groups.

3.2.2. Is the own-age effect evident in the patients while performing the FER?

To investigate whether the own-age effect exists in SCD, MCI, and AD groups,

and to evaluate the FER abilities in these groups, a mixed-effects ANCOVA with three

factors: group (between-subjects), photo age (within-subjects), and stimulus emotion

(within-subjects), was conducted. To control the possible effects of age, education,

and IQ, we included these factors as covariates. The dependent variable was the

proportion of correct classifications for each stimulus emotion in different photo ages

(i.e., younger and older faces). The results are shown in Table 5.

The results revealed a significant main effect of group (F(3, 100) = 7.34, p < .001,

ηp2 = .180) and significant two-way interactions of photo age ✕ group (F(3, 100) =

3.04, p = .033, ηp2 = .083) and emotion ✕ group (F(12, 400) = 2.12, p = .015, ηp2

= .060). A post hoc comparison using Scheffe’s method revealed that both HC and

SCD groups did significantly better than MCI and AD groups, while there was no

difference between HC and SCD as well as MCI and AD groups (see Figure 1). For

the interaction of photo age and group, further tests of simple main effect of photo age

showed that SCD group performed significantly better in decoding younger face than

in older face (F(1, 100) = 6.89, p = .001, ηp2 = .065). For the interaction of emotion

and group, further tests of simple main effect of group in decoding sadness indicated

that HC and SCD groups performed significantly better than MCI and AD groups (F(3,

403) = 20.86, p < .001, ηp2 = .134) in sadness expressions (see Figure 2), suggesting

that the accuracy difference between groups was mainly from the discrepancy in

decoding this category of expressions.

As the scores of sadness between groups could discriminate MCI and AD groups

from HC and SCD groups, it means that sadness recognition presents a remarkable

opportunity to discriminate patients who showed deficits in FER. Although the photo

ages did not show the main effect or interaction with the group in ANCOVA, we still

checked whether the own-age effect exists in sadness across groups to account for the

possibility that the own-age effect in sadness might be diminished by averaging total

emotion. We conducted a mixed-effects ANCOVA in sadness with two factors: photo

age (within-subjects) and group age (between-subjects). No significant interaction

between photo age and group (F(3, 100) = 1.261, p = .292) was found. However, the

results showed a trend that HC and SCD groups performed more accurately in

decoding younger faces compared to older faces (this trend also appeared for older

and younger observers in experiment 1), while MCI and AD groups performed more

accurately in decoding older faces than younger faces (see Figure 3). That is, it

seemed to show that the own-age trend existed in MCI and AD, but not HC and SCD.

To further explore which emotions tended to be mislabeled as from sadness by MCI

and AD groups, two-way mixed ANOVAs were conducted in these two groups.

Post-hoc analysis using the Bonferroni method found that the scores of judging

sadness to sadness were not significantly different from the scores of judging sadness

to anger and sadness to neutral. This means that MCI and AD groups tend to mislabel

sadness to either anger or neutral. To further examine whether there were differences

between mislabeling sadness as anger and as neutral under different photo ages

between the four groups, we conducted separate one-way ANCOVAs between the

four groups. The dependent variable was the proportion of wrong classifications from

sad to anger and neutral in younger and older faces, respectively. The results showed a

significant difference (F(3, 100) = 4.692, p = .004, ηp2 = .123) in mislabeling sadness

as anger in younger faces and a significant difference (F(3, 100) = 3.141, p = .029, ηp2

= .086) in mislabeling sadness as neutral in older faces across four groups. Post-hoc

analysis using the Bonferroni method confirmed that MCI and AD groups got higher

mislabeling scores compared to HC and SCD groups.

3.2.3. Clinical Utilities

From the above results, we thought sadness recognition presents a remarkable

opportunity to discriminate patients who showed deficits in FER. Thus, we conducted

the ROC curve analysis in sadness scores between SCD and MCI. The results

indicated that the sadness accuracies in younger and older faces were different

between SCD and MCI groups (area under the curve [AUC] of younger faces = 80%;

AUC of old faces = 77%). According to the Youden index (Youden, 1950), the data

showed that using a cutoff score of 0.35 for the accuracy in younger faces and a cutoff

score of 0.36 for the accuracy in older faces yielded the most desirable combination of

sensitivity (91%) and specificity (39%) in younger faces and sensitivity (81%) and

specificity (39%) in older faces respectively for identifying significant differences

between the SCD and MCI groups on the FER.