dat_pp <- subset(dat_long, cond == "posi" & metric == "具体性" & time %in% c("pre","post"))
dat_pp$group <- factor(dat_pp$group, levels = c("control","training"))
dat_pp$time <- factor(dat_pp$time, levels = c("pre","post"))
# 1) 线性混合模型
fit_pp <- lmer(score ~ group * time + (1|ID), data = dat_pp, REML = TRUE)
summary(fit_pp)
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: score ~ group * time + (1 | ID)
## Data: dat_pp
##
## REML criterion at convergence: 416.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.92915 -0.60060 -0.02392 0.56008 1.54255
##
## Random effects:
## Groups Name Variance Std.Dev.
## ID (Intercept) 0.9304 0.9646
## Residual 1.2784 1.1307
## Number of obs: 118, groups: ID, 59
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 2.9310 0.2760 96.8202 10.620 < 2e-16 ***
## grouptraining 0.1690 0.3870 96.8202 0.437 0.66340
## timepost 0.5862 0.2969 57.0000 1.974 0.05320 .
## grouptraining:timepost 1.3138 0.4164 57.0000 3.155 0.00256 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) grptrn timpst
## grouptranng -0.713
## timepost -0.538 0.384
## grptrnng:tm 0.384 -0.538 -0.713
# 2) 固定效应的 d（= estimate / sigma）—— 修正版
sig <- sigma(fit_pp)
co <- as.data.frame(summary(fit_pp)$coefficients)
fixed_d_pp <- tibble(
term = rownames(co),
estimate = co[,"Estimate"],
std.error = co[,"Std. Error"],
df = co[,"df"],
t = co[,"t value"],
p = co[,"Pr(>|t|)"]
) %>%
dplyr::mutate(
d = estimate / sig,
SE_d = std.error / sig,
d_low = d + qt(0.025, df) * SE_d,
d_high = d + qt(0.975, df) * SE_d
)
fixed_d_pp
## # A tibble: 4 × 10
## term estimate std.error df t p d SE_d d_low d_high
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 (Interce… 2.93 0.276 96.8 10.6 6.30e-18 2.59 0.244 2.11 3.08
## 2 grouptra… 0.169 0.387 96.8 0.437 6.63e- 1 0.149 0.342 -0.530 0.829
## 3 timepost 0.586 0.297 57.0 1.97 5.32e- 2 0.518 0.263 -0.00740 1.04
## 4 grouptra… 1.31 0.416 57.0 3.16 2.56e- 3 1.16 0.368 0.425 1.90
# 3) 组内简单效应（post − pre）+ d —— 如不需要可删掉
emm <- emmeans(fit_pp, ~ time | group)
w <- summary(contrast(emm, list(post_minus_pre = c(-1, 1)),
by = "group", adjust = "none"))
within_d <- as.data.frame(w) %>%
dplyr::mutate(
d = estimate / sig,
SE_d = SE / sig
) %>%
dplyr::mutate(
d_low = d + qt(0.025, df) * SE_d,
d_high = d + qt(0.975, df) * SE_d
)
within_d
## contrast group estimate SE df t.ratio p.value
## 1 post_minus_pre control 0.5862069 0.2969236 57 1.974268 5.320382e-02
## 2 post_minus_pre training 1.9000000 0.2919329 57 6.508344 2.088828e-08
## d SE_d d_low d_high
## 1 0.5184683 0.2626129 -0.007404917 1.044341
## 2 1.6804472 0.2581989 1.163412819 2.197482
dat_pf <- subset(dat_long, cond == "posi" & metric == "具体性" & time %in% c("pre","fu"))
dat_pf$group <- factor(dat_pf$group, levels = c("control","training"))
dat_pf$time <- factor(dat_pf$time, levels = c("pre","fu"))
# LMM
fit_pf <- lmer(score ~ group * time + (1 | ID), data = dat_pf, REML = TRUE)
summary(fit_pf)
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: score ~ group * time + (1 | ID)
## Data: dat_pf
##
## REML criterion at convergence: 417.5
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8436 -0.5957 0.0470 0.6314 1.8777
##
## Random effects:
## Groups Name Variance Std.Dev.
## ID (Intercept) 0.887 0.9418
## Residual 1.324 1.1507
## Number of obs: 118, groups: ID, 59
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 2.9310 0.2761 98.1983 10.615 < 2e-16 ***
## grouptraining 0.1690 0.3872 98.1983 0.436 0.66355
## timefu 0.9655 0.3022 57.0000 3.195 0.00228 **
## grouptraining:timefu 0.5555 0.4238 57.0000 1.311 0.19521
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) grptrn timefu
## grouptranng -0.713
## timefu -0.547 0.390
## grptrnng:tm 0.390 -0.547 -0.713
# 固定效应 d（= estimate / sigma）
sig_pf <- sigma(fit_pf)
co_pf <- as.data.frame(summary(fit_pf)$coefficients)
fixed_d_pf <- tibble::tibble(
term = rownames(co_pf),
estimate = co_pf[,"Estimate"],
std.error = co_pf[,"Std. Error"],
df = co_pf[,"df"],
t = co_pf[,"t value"],
p = co_pf[,"Pr(>|t|)"],
d = estimate / sig_pf,
SE_d = std.error / sig_pf,
d_low = d + qt(0.025, df) * SE_d,
d_high = d + qt(0.975, df) * SE_d
)
fixed_d_pf
## # A tibble: 4 × 10
## term estimate std.error df t p d SE_d d_low d_high
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 (Intercept) 2.93 0.276 98.2 10.6 5.51e-18 2.55 0.240 2.07 3.02
## 2 grouptrain… 0.169 0.387 98.2 0.436 6.64e- 1 0.147 0.337 -0.521 0.815
## 3 timefu 0.966 0.302 57.0 3.19 2.28e- 3 0.839 0.263 0.313 1.36
## 4 grouptrain… 0.555 0.424 57.0 1.31 1.95e- 1 0.483 0.368 -0.255 1.22
# 组内简单效应（fu − pre）及 d
emm_pf <- emmeans(fit_pf, ~ time | group)
w_pf <- summary(contrast(emm_pf, list(fu_minus_pre = c(-1, 1)),
by = "group", adjust = "none"))
within_d_pf <- w_pf %>%
as.data.frame() %>%
dplyr::mutate(
d = estimate / sig_pf,
SE_d = SE / sig_pf,
d_low = d + qt(0.025, df) * SE_d,
d_high = d + qt(0.975, df) * SE_d
)
within_d_pf
## contrast group estimate SE df t.ratio p.value d
## 1 fu_minus_pre control 0.9655172 0.3021992 57 3.194969 2.280038e-03 0.839040
## 2 fu_minus_pre training 1.5210000 0.2971199 57 5.119146 3.769345e-06 1.321758
## SE_d d_low d_high
## 1 0.2626129 0.3131668 1.364913
## 2 0.2581989 0.8047234 1.838792
dat_pfu <- subset(dat_long, cond == "posi" & metric == "具体性" & time %in% c("post","fu"))
dat_pfu$group <- factor(dat_pfu$group, levels = c("control","training"))
dat_pfu$time <- factor(dat_pfu$time, levels = c("post","fu"))
# LMM
fit_pfu <- lmer(score ~ group * time + (1 | ID), data = dat_pfu, REML = TRUE)
summary(fit_pfu)
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: score ~ group * time + (1 | ID)
## Data: dat_pfu
##
## REML criterion at convergence: 389.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5245 -0.4560 0.1945 0.5785 1.6985
##
## Random effects:
## Groups Name Variance Std.Dev.
## ID (Intercept) 0.7992 0.8940
## Residual 0.9794 0.9897
## Number of obs: 118, groups: ID, 59
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 3.5172 0.2476 94.8508 14.202 < 2e-16 ***
## grouptraining 1.4828 0.3473 94.8508 4.269 4.65e-05 ***
## timefu 0.3793 0.2599 57.0000 1.459 0.150
## grouptraining:timefu -0.7583 0.3645 57.0000 -2.081 0.042 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) grptrn timefu
## grouptranng -0.713
## timefu -0.525 0.374
## grptrnng:tm 0.374 -0.525 -0.713
# 如需 Type III： anova(fit_pfu, type = 3, ddf = "Kenward-Roger")
# 固定效应 d（= estimate / sigma）
sig_pfu <- sigma(fit_pfu)
co_pfu <- as.data.frame(summary(fit_pfu)$coefficients)
fixed_d_pfu <- tibble::tibble(
term = rownames(co_pfu),
estimate = co_pfu[,"Estimate"],
std.error = co_pfu[,"Std. Error"],
df = co_pfu[,"df"],
t = co_pfu[,"t value"],
p = co_pfu[,"Pr(>|t|)"],
d = estimate / sig_pfu,
SE_d = std.error / sig_pfu,
d_low = d + qt(0.025, df) * SE_d,
d_high = d + qt(0.975, df) * SE_d
)
fixed_d_pfu
## # A tibble: 4 × 10
## term estimate std.error df t p d SE_d d_low d_high
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 (Intercep… 3.52 0.248 94.9 14.2 3.27e-25 3.55 0.250 3.06 4.05
## 2 grouptrai… 1.48 0.347 94.9 4.27 4.65e- 5 1.50 0.351 0.802 2.19
## 3 timefu 0.379 0.260 57.0 1.46 1.50e- 1 0.383 0.263 -0.143 0.909
## 4 grouptrai… -0.758 0.364 57.0 -2.08 4.20e- 2 -0.766 0.368 -1.50 -0.0288
# 组内简单效应（fu − post）及 d
emm_pfu <- emmeans(fit_pfu, ~ time | group)
w_pfu <- summary(
contrast(emm_pfu, list(fu_minus_post = c(-1, 1)),
by = "group", adjust = "none")
)
within_d_pfu <- w_pfu %>%
as.data.frame() %>%
dplyr::mutate(
d = estimate / sig_pfu,
SE_d = SE / sig_pfu,
d_low = d + qt(0.025, df) * SE_d,
d_high = d + qt(0.975, df) * SE_d
)
within_d_pfu
## contrast group estimate SE df t.ratio p.value d
## 1 fu_minus_post control 0.3793103 0.2598981 57 1.459458 0.1499286 0.3832724
## 2 fu_minus_post training -0.3790000 0.2555298 57 -1.483193 0.1435317 -0.3829588
## SE_d d_low d_high
## 1 0.2626129 -0.1426008 0.9091456
## 2 0.2581989 -0.8999932 0.1340756
dat_pp2 <- subset(dat_long, cond=="posi" & metric=="可能性" & time %in% c("pre","post"))
dat_pp2$group <- factor(dat_pp2$group, levels=c("control","training")) # ← 用 dat_pp2
dat_pp2$time <- factor(dat_pp2$time, levels=c("pre","post"))
# 1) 模型
fit_pp2 <- lmer(score ~ group * time + (1|ID), data = dat_pp2, REML = TRUE)
summary(fit_pp2)
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: score ~ group * time + (1 | ID)
## Data: dat_pp2
##
## REML criterion at convergence: 294
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.54184 -0.52023 0.04243 0.46514 2.17333
##
## Random effects:
## Groups Name Variance Std.Dev.
## ID (Intercept) 0.1778 0.4217
## Residual 0.5302 0.7281
## Number of obs: 118, groups: ID, 59
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 6.9703 0.1562 107.2345 44.610 < 2e-16 ***
## grouptraining -0.3393 0.2191 107.2345 -1.549 0.12441
## timepost -0.1079 0.1912 57.0000 -0.564 0.57466
## grouptraining:timepost 0.7486 0.2682 57.0000 2.792 0.00712 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) grptrn timpst
## grouptranng -0.713
## timepost -0.612 0.436
## grptrnng:tm 0.436 -0.612 -0.713
# 2) 固定效应 d
sig <- sigma(fit_pp2)
co <- as.data.frame(coef(summary(fit_pp2)))
fixed_d_pp2 <- tibble::rownames_to_column(co, "term") |>
dplyr::transmute(
term,
estimate = Estimate,
std.error = `Std. Error`,
df,
t = `t value`,
p = `Pr(>|t|)`,
d = estimate / sig
)
fixed_d_pp2
## term estimate std.error df t p
## 1 (Intercept) 6.9703448 0.1562494 107.2345 44.610366 4.564472e-71
## 2 grouptraining -0.3393448 0.2191209 107.2345 -1.548665 1.244087e-01
## 3 timepost -0.1079310 0.1912152 57.0000 -0.564448 5.746639e-01
## 4 grouptraining:timepost 0.7485977 0.2681562 57.0000 2.791648 7.123765e-03
## d
## 1 9.5729949
## 2 -0.4660525
## 3 -0.1482313
## 4 1.0281159
# 3) 组内（post − pre）效应量 d
emm <- emmeans(fit_pp2, ~ time | group)
w <- summary(contrast(emm, list(post_minus_pre = c(-1, 1)),
by = "group", adjust = "none"))
within_d_pp2 <- as.data.frame(w) |>
dplyr::mutate(
d = estimate / sig,
SE_d = SE / sig
) |>
dplyr::mutate(
d_low = d + qt(0.025, df) * SE_d,
d_high = d + qt(0.975, df) * SE_d
)
within_d_pp2
## contrast group estimate SE df t.ratio p.value
## 1 post_minus_pre control -0.1079310 0.1912152 57 -0.564448 0.574663872
## 2 post_minus_pre training 0.6406667 0.1880013 57 3.407778 0.001207626
## d SE_d d_low d_high
## 1 -0.1482313 0.2626129 -0.6741045 0.3776419
## 2 0.8798846 0.2581989 0.3628502 1.3969189