1
$\begingroup$

I need to compare total concentrations of a chemical across different tissues of a plant species. I sampled multiple tissues from each individual, and sampled at multiple sites. I do not know how to handle the individual and site covariates. There is good reason to expect a site effect on chemical concentrations. I need to control for the fact that concentrations will be correlated across tissues within individuals, but am only interested in whether concentrations differ consistently across tissues. I.e. are leaf concentrations always the lowest, etc.

I have two questions -

Question 1 what it the most appropriate analysis to use?

I do not think I have enough data for a mixed effects model, and would like to avoid using one if possible.

So far, I've used: (1) a three way ANOVA with tissue, individual, and site as factors.

3wy_mod <- (lm(total_conc ~ tissue+indv+site, data=data))

However, I feel like that isn't correct. I think that the model structure should have tissue nested within individual nested within site. (2) to address the nesting issue, I've also used this model:

nest <- anova(lm(total_conc ~ tissue/indv/site, data=data))

The nested model seems correct. However, I'm not sure how to validate this model. This brings me to

Question 2

How do I assess model fit, check assumptions, etc with whichever model structure turns out to be appropriate?

Here is an attempt at a reproducible example:

data <- structure(list(indv = c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 3L, 3L, 
3L, 3L, 4L, 4L, 4L, 4L, 5L, 5L, 5L, 5L, 6L, 6L, 6L, 6L, 7L, 7L, 
7L, 8L, 8L, 8L, 8L, 9L, 9L, 9L, 9L, 10L, 10L, 10L, 10L, 11L, 
11L, 11L, 11L, 12L, 12L, 12L, 12L, 13L, 13L, 13L, 13L, 14L, 14L, 
14L, 14L, 15L, 15L, 15L, 15L, 16L, 16L, 16L, 16L, 17L, 17L, 17L, 
17L, 18L, 18L, 18L, 18L, 19L, 19L, 19L, 19L, 20L, 20L, 20L, 20L, 
21L, 21L, 21L, 21L, 22L, 22L, 22L, 22L, 23L, 23L, 23L, 23L, 24L, 
24L, 24L, 24L, 25L, 25L, 25L, 25L, 25L, 26L, 26L, 26L, 26L, 26L, 
27L, 27L, 27L, 27L, 28L, 28L, 28L, 28L), tissue = structure(c(3L, 
5L, 1L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 1L, 4L, 3L, 
5L, 1L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 1L, 3L, 1L, 5L, 4L, 3L, 1L, 
5L, 4L, 3L, 1L, 5L, 4L, 3L, 1L, 5L, 4L, 3L, 1L, 5L, 4L, 3L, 1L, 
5L, 4L, 3L, 1L, 5L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 
1L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 1L, 4L, 3L, 5L, 
1L, 4L, 3L, 1L, 5L, 2L, 3L, 1L, 5L, 4L, 3L, 1L, 5L, 4L, 4L, 3L, 
1L, 5L, 4L, 3L, 1L, 5L, 2L, 4L, 3L, 1L, 5L, 4L, 3L, 1L, 5L, 4L
), .Label = c("flower", "fruit", "leaf", "pollen", "stem"), class =          "factor"),
site = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 
3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 4L, 4L, 
4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 
4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L), .Label = c("a", 
"b", "c", "d"), class = "factor"), total_conc = c(8251.26, 
3845.59, 6681.23, 504.47, 21494.77, 13514.46, 14244.08, 1091, 
4156.25, 8301.19, 12968.21, 5536.14, 9034.05, 11401.89, 12010.11, 
2013.21, 14997.58, 12266.25, 23068.28, 4784.39, 9326.1, 5522.93, 
9814.8, 3467.77, 7107.22, 6129.18, 7330.3, 7242.1, 8668.38, 
5214.82, 1891.02, 5918.55, 11268.32, 3267.63, 3648.9, 13406.13, 
15847.31, 6775.97, 1781.21, 10943.45, 14477.16, 7890.34, 
1738.36, 14467.25, 11777.99, 5551.34, 1136, 5537.86, 18114.18, 
2895.41, 999.91, 3633.19, 7243.96, 4856.73, 8317.76, 15885.53, 
6811.57, 15818.11, 1316.81, 15347.71, 8068.19, 14688.62, 
2247.32, 17762.4, 6628.74, 16494.26, 11255.5, 19459.68, 5960.92, 
20607.73, 3226.06, 11572.8, 7924.07, 12091.07, 2934.25, 12734.17, 
5158.48, 16520.8, 1431.31, 14007.17, 4689.99, 23774.68, 5799.91, 
12903.34, 18982.81, 9125.87, 14869.07, 20022.23, 28254.47, 
9459.41, 1977.36, 22562.52, 22195.51, 12376.81, 2284.3, 1358.63, 
24866.6, 15789.27, 9544.07, 275.32, 17669.95, 18935.79, 9357.06, 
18818.68, 1676.3, 23224.24, 25432.66, 12328.59, 1367.51, 
24176.56, 26710.84, 21184.79, 1642.89)), .Names = c("indv", 
"tissue", "site", "total_conc"), class = "data.frame", row.names = c(NA, 
-113L))

3wy_mod <- anova(lm(total_conc ~ tissue+indv+site, data=data))

nest <- anova(lm(total_conc ~ tissue/indv/site, data=data))
$\endgroup$
  • $\begingroup$ A few questions before an answer: (1) Are individuals only at one site? Or can an individual come from multiple sites? (2) You mention not having enough data. How many sites are there? How many individuals do you have, per site (on average), and how many tissues do you have, per person (on average)? $\endgroup$ – Mark White Nov 2 '17 at 16:25
  • $\begingroup$ One more: (3) What are your predictor variables of interest? And what are they characteristics of? Are they characteristics of the tissue, the individual, or the site? Or do you have predictors and covariates at all levels? $\endgroup$ – Mark White Nov 2 '17 at 16:26
  • 1
    $\begingroup$ I agree with @EdM . I also don't see why you say you don't have enough data for a mixed model? Either lmer(total_conc ~ tissue + (1|indv) + (1|site), data), or total_conc ~ tissue + site (1|indv), data works fine for me. $\endgroup$ – Stefan Nov 2 '17 at 21:04
  • 1
    $\begingroup$ Maybe in that case you don't have enough data to compute the Satterthwaite's approximation for degrees of freedom (but I don't know that). Have a look at the afex package and the mixed() function. Then look at the methods argument where you can specify different methods for obtaining p-values. $\endgroup$ – Stefan Nov 2 '17 at 21:52
  • 1
    $\begingroup$ @Stefan @EdM @JKO it doesn't seem like there would be enough data for a three-level multilevel model, imo. Just because it runs doesn't mean you are getting valid estimates, right? It isn't a perfect solution, but I would suggest doing a Bayesian multilevel model. Even weakly informative priors could nudge the estimation process along. If you don't know Stan, but know lme4, the brms package specifies models just like lme4, but has extra arguments for priors. You need a C++ compiler for it, but it's a fantastic package for multilevel models. $\endgroup$ – Mark White Nov 2 '17 at 22:54
2
$\begingroup$

The crossed, non-nested model is incorrect, as you suspect, and seems to be over-specified for your posted data. It's important to get the nesting correct, which can be confusing. This page is a good guide. With multiple individuals per site and each individual restricted to one site, the individuals are nested within site. If you were to use your anova(lm()) approach you should be writing:

total_conc ~ tissue/site/indiv

which is expanded to:

total_conc ~ 1 + tissue + tissue:site + tissue:site:indiv

So this model examines the main effect of tissue plus its interactions with site and with site:indiv, without main effects for site or indiv. The result of your anova(lm()) on your posted data* would then be:

Analysis of Variance Table

Response: total_conc
                 Df     Sum Sq   Mean Sq  F value  Pr(>F)  
tissue            4 3023661191 755915298 1288.243 0.02089 *
tissue:site      12 1280847591 106737299  181.903 0.05788 .
tissue:site:indv 95 1325836886  13956178   23.784 0.16203  
Residuals         1     586780    586780                   

You only have 1 residual degree of freedom because most cases have only one measurement on each tissue from an individual. So the three-way interaction is fairly meaningless. Although results from the Type I sums of squares used by anova(), when applied to unbalanced data, can depend on the order of entry of variables into the model formula, in this case there really is only one order of entry that makes sense, so that's not such an issue.

These data are highly unbalanced, with observations on fruit restricted to 2 indiv at one particular site. If that's characteristic of your real data, then you probably should consider removing such tissues with low numbers of observations from your analyses. Would you (or a skeptical reviewer) really trust analyses based on 2 observations from a single site? Removing the fruit observations from these data, at least, would leave a reasonably balanced design that would probably be good enough for standard nested ANOVA.

Note that both the nested ANOVA approach and mixed models with intercept-only terms like (1|indiv) are making an implicit assumption that all effects of individuals and sites on total_conc are additive without regard to the tissue being evaluated. You must use your knowledge of the subject matter as to the validity of that assumption. With the raw means of total_conc ranging from 2804 (pollen) to 16844 (fruit) I would be a bit worried about that assumption. In principle the formalism of a mixed model could incorporate random effects that differ among tissue values, but you don't seem to have enough data for that. The suggestion from @MarkWhite to try a Bayesian multilevel model might help, but I have no experience with such models.

If there is some irreducible imbalance in the data then I think that many would argue for a mixed-model approach. My guess is that the errors that you discover with mixed models due to small numbers of observations would also be seen in the anova(lm()) approach if you dug more deeply. For example, that approach on the crossed model applied to your data provides what seems to be a perfectly reasonable ANOVA table.

> anova(lm(total_conc ~ tissue+site+indv, data=data))
Analysis of Variance Table

Response: total_conc
          Df     Sum Sq   Mean Sq F value    Pr(>F)    
tissue     4 3023661191 755915298  47.967 < 2.2e-16 ***
site       3  746832764 248944255  15.797 3.552e-08 ***
indv      24  583953341  24331389   1.544   0.07733 .  
Residuals 81 1276485152  15759076                      

Nevertheless, the underlying crossed model can't define 3 regression coefficients because of singularities:

> summary(lm(total_conc ~ tissue+site+indv, data=data))

Call:
lm(formula = total_conc ~ tissue + site + indv, data = data)

Residuals:
    Min      1Q  Median      3Q     Max 
-9199.3 -2226.1   -84.8  1890.3  9891.5 

Coefficients: (3 not defined because of singularities)

*Note that the indiv values in the posted data need to be converted to factors first.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Many thanks for your answer, it addressed my question exactly and thoroughly. I have indeed dropped fruit from all analyses. $\endgroup$ – JKO Nov 4 '17 at 22:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.