Table 1. Cohesion predicts community turnover in six long-term time series.

Lake	Taxon pers. cutoff a	Adjusted R2	Positive cohesion P-value	Negative cohesion P-value	Positive cohesion direction b	Negative cohesion direction b	Days between samples	Number of samples
Mendota (phyto)	5%	0.465	0.405	<1 × 10−20	NS	Stronger is stabilizing	36–48	186
Monona	5%	0.355	0.413	<1 × 10−15	NS	Stronger is stabilizing	36–48	166
Peter	10%	0.357	0.062	<1 × 10−3	NS	Stronger is stabilizing	39–45	121
Paul	10%	0.500	<1 × 10−11	<1 × 10−19	Weaker is stabilizing	Stronger is stabilizing	39–45	125
Tuesday	10%	0.374	0.355	<1 × 10−8	NS	Stronger is stabilizing	39–45	72
Mendota (16S)	5%	0.378	0.0039	<1 × 10−5	Weaker is stabilizing	Stronger is stabilizing	25–41	54

^aStands for ‘taxon persistence cutoff’, which was the minimum proportion of presences across the data set that we used as a cutoff for including taxa in the connectedness and cohesion metrics. Other cutoffs may give higher model-adjusted R² values (see the Supplementary Online Material), but we wanted to use the same cutoff for data sets collected within the same sampling program. We also applied a mean abundance cutoff to the Lake Mendota 16S rRNA gene data set, where we removed taxa with a mean abundance <1 × 10⁻⁷.

^bThese columns indicate the direction of a significant relationship between cohesion and Bray–Curtis dissimilarity. For example, ‘stronger is stabilizing’ means that greater cohesion is related to lower Bray–Curtis dissimilarity. Nonsignificant relationships are denoted ‘NS’.