Manchester Hardening project growth and survival analysis
Overview
Over the last few weeks students in the lab have been analyzing images to measure oyster growth from the Manchester Hardening experiment. This post shows my initial analysis of those data.
All data are on GitHub here.
Scripts to analyse these data are also on GitHub here.
Growth and survival were measured on the following time points:
- 2025-05-31: initial measurement following 2 week hardening
- 2025-09-23: growth and survival after first summer
- 2026-05-28: growth and survival after first full year
- 2026-08-14: growth and survival after second summer
For each family, there are two tags of Control oysters (no hardening) and two tags of High oysters (hardening treatment) with four total tags per family. There are 10 total families.
Survival analysis
Survival data is available on GitHub here.
We counted live oysters at each time point or obtained counts from images. At each time point after the start of the deployment, we have multiple tags with more oysters than were present at the start. I looked back at all of my notes and images and the counts are accurate. We must have had some recruitment into the bags perhaps. I need to look back and see if we can see a clear size difference to indicate potential recruits that shouldn’t be included in survival.
I ran a model with family and treatment as main effects, but not the interaction because we only have n=2 per treatment x family combination. We have n=4 per family and n=20 per treatment total.
Here is the full dataset - note the 100+% survival observations.
Here is survival by family.
And finally, here is survival by treatment.
Statistical results show an effect of family, but not treatment.
model2<-data%>%
lmer(survival~treatment+family+as.factor(date)+treatment:as.factor(date)+family:as.factor(date)+ (1|cage) + (1|tank), data=.)
anova(model2)
qqPlot(residuals(model2))
emm<-emmeans(model2, ~family | date)
pairs(emm)
cld(emm)There is an effect of family but not treatment.
Type II Analysis of Variance Table with Satterthwaite's method
Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
treatment 25.2 25.20 1 2 0.4012 0.5912413
family 4174.7 463.86 9 114 7.3844 1.977e-08 ***
as.factor(date) 2241.5 747.17 3 114 11.8945 7.799e-07 ***
treatment:as.factor(date) 421.4 140.48 3 114 2.2364 0.0878096 .
family:as.factor(date) 4001.5 148.20 27 114 2.3593 0.0008999 ***Post hoc results show that survival wasn’t different after the first summer. Survival begins to differentiate between families a little by one year post deployment, but differentiate much more at the end of the deployment (end of the second summer).
date = 20250531:
family emmean SE df lower.CL upper.CL .group
YC24-163 100.0 4.17 72.4 91.7 108.3 1
YC24-153 100.0 4.17 72.4 91.7 108.3 1
YC24-116 100.0 4.17 72.4 91.7 108.3 1
YC24-164 100.0 4.17 72.4 91.7 108.3 1
YC24-099 100.0 4.17 72.4 91.7 108.3 1
YC24-137 100.0 4.17 72.4 91.7 108.3 1
YC24-148 100.0 4.17 72.4 91.7 108.3 1
YC24-105 100.0 4.17 72.4 91.7 108.3 1
YC24-158 100.0 4.17 72.4 91.7 108.3 1
YC24-087 100.0 4.17 72.4 91.7 108.3 1
date = 20250923:
family emmean SE df lower.CL upper.CL .group
YC24-153 95.6 4.17 72.4 87.2 103.9 1
YC24-087 100.7 4.17 72.4 92.4 109.0 1
YC24-163 100.9 4.17 72.4 92.6 109.2 1
YC24-164 101.8 4.17 72.4 93.5 110.1 1
YC24-148 105.4 4.17 72.4 97.1 113.7 1
YC24-105 106.9 4.17 72.4 98.6 115.3 1
YC24-158 107.4 4.17 72.4 99.0 115.7 1
YC24-099 110.0 4.17 72.4 101.7 118.3 1
YC24-116 110.3 4.17 72.4 102.0 118.6 1
YC24-137 112.5 4.17 72.4 104.2 120.8 1
date = 20260528:
family emmean SE df lower.CL upper.CL .group
YC24-153 84.4 4.17 72.4 76.1 92.8 1
YC24-105 90.3 4.17 72.4 82.0 98.6 12
YC24-087 95.4 4.17 72.4 87.1 103.7 12
YC24-148 96.4 4.17 72.4 88.1 104.7 12
YC24-163 97.3 4.17 72.4 89.0 105.6 12
YC24-099 97.5 4.17 72.4 89.2 105.8 12
YC24-158 98.5 4.17 72.4 90.2 106.8 12
YC24-164 100.0 4.17 72.4 91.7 108.3 12
YC24-116 107.7 4.17 72.4 99.4 116.0 2
YC24-137 107.8 4.17 72.4 99.5 116.1 2
date = 20260814:
family emmean SE df lower.CL upper.CL .group
YC24-153 77.2 4.17 72.4 68.9 85.5 1
YC24-087 80.3 4.17 72.4 72.0 88.6 12
YC24-163 86.6 4.17 72.4 78.3 94.9 123
YC24-116 87.5 4.17 72.4 79.2 95.8 123
YC24-148 96.4 4.17 72.4 88.1 104.7 234
YC24-099 97.5 4.17 72.4 89.2 105.8 234
YC24-164 100.0 4.17 72.4 91.7 108.3 345
YC24-105 101.4 4.17 72.4 93.1 109.7 345
YC24-158 105.9 4.17 72.4 97.6 114.2 45
YC24-137 117.2 4.17 72.4 108.9 125.5 5Growth analysis
Growth data is available on GitHub here.
At each time point, we photographed oysters and then analyzed size in ImageJ. Christina, Maddy, and Grace were a huge help in analyzing the images!
For this dataset, I am analyzing size over time so that I can use all observations. I did not collapse growth to an average rate for each tag, because that will severely limit our sample size.
Oyster volume was analyzed using the following model:
calc_oyster_volume <- function(length_mm, width_mm, k = 0.21) {
k * length_mm * (width_mm^2)
}
data$volume.mm3 <- calc_oyster_volume(data$length.mm, data$width.mm)I analyzed size over time with treatment, time, and family as main effects and tag and hardening tank as random effects. Time is shown as days post deployment.
Here are the size data for all of the replicate tags.
Here are the size data summarized by family and treatment.
After running the models (below) I conducted a quarter-root transformation. Here are the data displayed on the same scale as included in the model to compare the results too.
model<-data%>%
lmer((volume.mm3)^(1/4) ~ as.factor(days) * family * treatment + (1|tag) + (1|cage), data=.)
summary(model)
anova(model, type=2)
emm <- emmeans(model, ~ as.factor(days) * family * treatment)
pairs(emm, by = c("days", "family"), adjust = "tukey")
qqPlot(residuals(model))Type II Analysis of Variance Table with Satterthwaite's method
Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
as.factor(days) 25246.4 8415.5 3 3045.25 3766.1173 < 2.2e-16 ***
family 560.2 62.2 9 15.68 27.8550 5.206e-08 ***
treatment 0.4 0.4 1 2.21 0.1672 0.7189
as.factor(days):family 1089.1 41.9 26 3044.29 18.7465 < 2.2e-16 ***
as.factor(days):treatment 6.3 2.1 3 3046.30 0.9408 0.4200
family:treatment 13.2 1.5 9 15.70 0.6586 0.7333
as.factor(days):family:treatment 153.9 5.9 26 3045.06 2.6488 1.142e-05 ***There is a significant interaction between date, family, and treatment. I then ran a post hoc to determine which family x treatment effects are present.
- YC24-163 on days 116 P=0.043 (smaller size in treated)
- YC24-153 on days 116 P=0.041 (smaller size in treated)
- YC24-137 on days 116 P=0.001 (bigger size in treated)
- YC24-105 on days 116 P=0.001 (bigger size in treated)
Day 116 was the September 2025 time point, representing effects of hardening after the first summer of deployment.
Assessing family specific effects
I then looked back at the acute stress survival data detailed in my past post. Here are the families of interest.
Hardening led to the following effects:
- Family YC-163 lost thermal tolerance and had smaller sizes after the 1st summer after hardening
- Family YC-153 gained thermal tolerance and had smaller sizes after the 1st summer after hardening
- Family YC-137 gained thermal tolerance and had larger sizes after the 1st summer after hardening
- Family YC-116 lost thermal tolerance, no size differences
- Family YC-105 larger sizes after the 1st summer after hardening
It is interesting to note that hardening effects are very family specific. In some cases, families that gained thermal tolerance also increased their growth after the 1st summer, while others did the opposite or weren’t affected.