{"id":106,"date":"2025-05-12T09:53:10","date_gmt":"2025-05-12T13:53:10","guid":{"rendered":"https:\/\/health.uconn.edu\/causality\/?page_id=106"},"modified":"2026-02-16T16:21:17","modified_gmt":"2026-02-16T21:21:17","slug":"aging","status":"publish","type":"page","link":"https:\/\/health.uconn.edu\/causality\/aging\/","title":{"rendered":"Causality challenges in aging research"},"content":{"rendered":"

*** Aging research invites some unique statistical modeling challenges: the natural course of development in the later years of life is generally marked by declines across the board. That\u2019s why many constructs describing such processes contain in their very cores this \u2018going down\u2019 direction of changes, like \u2018sarcopenia\u2019 (i.e. \u2018loss<\/em> of sarcomeres\u2019, and hence of muscle mass\/strength). Specialized methods have grown out of this \u2018data design\u2019, in which all cases (patients, residents, older adults in general) show declines, and there is a sharp end it too[i]<\/span><\/a>. Some topics:<\/p>\n

1. <\/strong>Some notational basics specific to causality in the context of \u2018changes\u2019:<\/p>\n

*** In the notation shown in step 1 (superscript<\/sup><\/strong> represents \u2018up there\u2019, in \u2018potential outcomes\u2019 world, while superscript<\/sub><\/strong> is \u2018down on earth, realized\u2019), common RQs related to aging are comparisons of potential outcome of \u2018frailty\u2019 (in an old person i<\/sub><\/em><\/strong> )<\/p>\n

Frailty <\/sub>i<\/sub><\/em><\/strong> <\/sup>10Years.Later<\/sup><\/em> vs. Frailty <\/sub>i<\/sub><\/em><\/strong> Now<\/sub><\/strong><\/p>\n

which asks how much would time<\/em> normally \u2018lead to\u2019 my level of frailty 10 years from now (what time is, is not a simple question btw[ii]<\/span><\/a>). Notice that this \u2018time effect\u2019 is different than the effect of some anti-aging multivitamins, to take an example:<\/p>\n

Frailtyi<\/sub> <\/sup>10Years.Later.IF.Multivitamins<\/strong><\/sup><\/em> vs. Frailty <\/sub>i<\/sub><\/em><\/strong> <\/sup>10Years.Later. If.RegularBusiness<\/sup><\/em><\/p>\n

* Here both are potential\/unknownable, because they both will happen in the future. This makes evident the two layers of causal questions (\u2018nested counter-factuals\u2019), one asking about \u2018the effect of time\u2019 (aging per se) vs. the \u2018derailing\u2019 effect of some \u2018aging prevention\u2019 initiative.<\/p>\n

2.<\/strong> How (much) time and causality overlap[iii]<\/a> ; the graphical window into causality appears to include time in the very definition of the directional edges in the graph[iv]<\/a>; time \u2018precedence\u2019 is often listed as a condition of causality[v]<\/a> (it may be the distinguishing attribute between \u2018cause\u2019 and \u2018effect\u2019 when the cause is a necessary and sufficient condition of the effect[vi]<\/a>).<\/p>\n

2.i. <\/strong>The definitions of \u2018time\u2019 and \u2018causality\u2019 overlap dangerously; and time itself is a difficult construct to spell out[vii]<\/span><\/a>, so much so that some scientists argue it may not exist at all (Russell Barbour<\/a>, e.g.[viii]<\/span><\/a>).<\/p>\n

2.ii. <\/strong>An added challenge is \u201cdifferentiating between true within-individual change and measurement error\u201d (see Lauren Gaydosh’s blog<\/a>)<\/p>\n

*** The conundrum is \u2018simple\u2019: if we measure A1c now, say 5.6, and tomorrow, say 5.9, is the 0.3 difference the \u2018true change\u2019? A 2% measurement uncertainty widens this change anywhere between a minimum of [(5.9+2%) -(5.6-2%)] = 0.07 &\u00a0 and a maximum of [(5.9=-2%) (5.6+2%)] =\u00a0 0.53, which is 7+ times larger. [ix]<\/span><\/a><\/p>\n

2.iii. <\/strong>On an other hand, reliability and repeated measurement overlap conceptually too[x]<\/span><\/a>.<\/p>\n

Some further topics to consider:<\/em><\/p>\n

+1. <\/em>Continuous time vs. discrete observation timepoints<\/p>\n

+2. Causal implications of time precision and the discrete vs. continuous time modeling<\/p>\n

+3. Differential missingness implications for causal conclusions<\/p>\n

– Recover causal effects from \u2018apparent\u2019 effects, paradoxical at times, e.g. apparent \u2018benefit\u2019 of Hispanic residents in the US in Covid mortality: age of the living mediates in fact the effect of ethnicity on Covid mortality: the \u2018true effect\u2019 is not the ethnicity effect controlled for age (\u2018age adjusted Covid mortality\u2019), but the \u2018total\u2019 effect, both through \u2018age\u2019 and directly.<\/p>\n

<\/strong>5. How to disentangle \u2018normal aging\u2019 changes from \u2018accelerated\/beyond-normal\u2019 aging, hence distinguishing aging disparities<\/p>\n

*** The next part, # 6 Tinyurl.com\/GEOCAUSAL<\/a> will go over some challenges and opportunities for raised by spatial data, similarities with over-time and dyadic (\u2018dependent\u2019) data, and some causal solutions provided by the \u2018ancient\u2019 path analytic tradition.<\/p>\n

FOOTnotes<\/strong><\/span><\/p>\n

[i]<\/span><\/a> This \u2018necessary end\u2019 point is evident in the calculations of life expectancy (at birth, e.g.) from mortality tables: from the percentages of deaths within each age group, e.g. how many 60 year olds died this year, tables of such probabilities of death are derived, by age, but this probability does not \u2018trail off\u2019 continuously upwardly, even though (very few) older adults can live even beyond 100 years of age, but becomes a sharp 1.0 for the \u2018last age group\u2019: everyone in that group will die, with probability 1 or 100% (the calculator has to impose this \u2018constraint\u2019, see \u00a0[1]. I show an adapted view of a figure <\/a>from [2], which shows how we need to \u2018force\u2019 everyone to die, even though we don\u2019t observe that outcome for them in the \u2018time window\u2019 the data covers.<\/p>\n

+++ This same view\u2019 is seen when one builds up a \u2018survival step graph\u2019, or curve, by hand, which shows the percent still \u2018living\u2019\u00a0 as a function of time (even though survival can be used for any process that has a clear end, like \u2018treatment adherence, until dropping out of treatment) \u2013 this curve has to come down and in the case of biological death has to touch the bottom, i.e. the horizontal axis, where the percent remaining living is a sharp 0. This illustrates also the peculiarity of age as a variable, which is bounded on left at 0, and on the right at some \u2018ceiling\u2019 value: the continuous unending distribution of values assumed in many statistical tests is not applicable to age.<\/p>\n

[ii]<\/span><\/a> Some famous phrasing put time in the center of metal controversies, e.g.: 1. \u201cTime is what prevents everything from happening at once\u201d; \u201ctime flows at a speed of 1 second per second\u201d. A more interesting inquiry looks into how come time is one-dimensional, but space is more than 1 dimensional; Van Fraassen touches upon this \u201cthe temporal relations give rise to a temporal order and the spatial relations to a spatial order. Of the two orders, the latter is by far more complex. Let us assume for a moment that between is indeed a basic relation for both time and space. Then we find nevertheless that the order to which temporal betweenness gives rise is much simpler than the order determined by the spatial betweenness relation. To mention the most obvious case: if X, Y, and Z are in time, but not simultaneous, then one of the three is between the other two; it is not true, however, that if X, Y, and Z are in space, though not in the same place, then one of them is between the other two. The reader may recognize this point in the form of the statement \u201cSpace is three-\u00addimensional, but time is only one-\u00ad\u2010dimensional.\u201d, [3] p. 11 & \u201cIn the Theodicy, Leibniz says that in geometry we can prove that there are only three straight lines perpendicular to one another that can intersect at one and the same point and that this shows that space has necessarily exactly three dimensions\u201d, p. 149, but more detailed \u2018how come\u2019 can be extracted from AIs, e.g. Claude.ai: “The key is that with four points in 3D space, you have exactly the right number of constraints. Each point has 3 coordinates, giving 12 parameters total. But you can eliminate rigid motions (3 translations + 3 rotations), leaving 6 degrees of freedom. The 6 distances between 4 points provide exactly these 6 constraints.”, but more detailed insights (\u2018break down\u2026\u2019) can then be asked for.<\/p>\n

[iii]<\/span><\/a> \u201cIt is yet unclear how the perception of simultaneity and successiveness relates to the present moment and event causality. Desantis and Buehner discuss the relation between time perception and causality. \u00a0Traditionally, it has been considered that causality is partly inferred from temporal relations between events with people being more likely to report two events as causally linked if they follow each other closely in time rather than if they are separated by a long delay. Desantis and Buehner review and discuss a complementary literature proposing that time perception is in turn guided by people\u2019s assumptions about the causal \u00a0connection between events. This implies that time perception is shaped by the combination of sensory and internal information such as our prior causal expectations.<\/p>\n

Finally, a discussion is devoted to how Bayesian models of uncertainty reduction may offer an exploratory hypothesis to understand how causality structures time perception.\u201d [4] p. xxvii<\/p>\n

*** Note that some constructs contain in themselves change\/dynamic conceptual attributes, beginning with the very \u2018aging\u2019 one (which requires >1 age!, and can go 1 step higher too: \u2018accelerated aging\u2019), but going then to: longevity<\/strong>, dysregulation<\/strong>, resilience<\/strong>, sarcopenia, increased<\/strong> risk, accumulation<\/strong> of cellular damage, impaired<\/strong> function, loss<\/strong> of physiological integrity, and more. Also of note that language tends to trick us, like using \u2018preventing aging\u2019.<\/p>\n

[iv]<\/span><\/a> \u201cThe most familiar connection between causality and conditional independence is reflected in the scientific notion of a state, which was devised to nullify the influence that the past exerts on the future by providing a\u00a0 sufficiently detailed description of the present. In probabilistic terms this came to be known as a Markov property: Future events are conditionally independent of past events, given the current state of affairs. This is precisely the role played by the set of parents \u041fX<\/sub><\/strong> in the construction of Bayesian networks (Section 3.3); they screen each variable X from the influence of all other predecessors of X.\u201d [5] P. 385 \u20188. Learning Structure from Data\\ 8.1. Causality, Modularity, and Tree Structures\u2019;<\/p>\n

\u201cOur probabilistic treatment of causation is influenced by that of Simon [[6]1954, [7]1980], and it stresses non-temporal causal ordering. Suppes [[8]1970] provides a probabilistic account of causation which is based on temporal ordering. A logic-based account of causation and temporal reasoning is\u00a0 given in Shoham [[9]1988]. A contrast between two conceptions of causal ordering, in the context of representing device behavior in AI systems, is described by Iwasaki and Simon ([10]1986) and de Kleer and Brown ([11]1986 [see [12]] too).\u201d P. 408<\/p>\n

[v]<\/span><\/a> \u201cIf, then, it be contended, in accordance with Mill\u2019s first position, that by a cause we mean a necessary and sufficient condition (more particularly, a necessary and sufficient precedent condition), it will appear that a hypothesis of causality really involves two universal propositions, and, it may be said, requires a double verification.<\/p>\n

This is what is implied in Mill’s account of the simplest use of the direct Method of Difference, where we first observe the absence of the factor in question as well as of the phenomenon it is :supposed to cause, this verifying the supposition that the factor is necessary, and then observe the phenomenon ensuing upon the introduction of the factor, this verifying the supposition that it is sufficient.\u201d [13] p. 132 [Mill (System of Logic, Bk. III, ch. X, \u00a7 3)]& \u201cIn logical form, if A is something specific ( and, if it is not, no assertion is being made), the assertion “All B are A” is equivalent to the assertion “All non-A are non-B”, each being the contrapositive of the other. What is brought out by the consideration that there may not be “negative instances” is that we are concerned not with relations between A and B in general but with their relations within certain limits or in a certain “field”; and it is the consideration of the “field” that enables us to make the theory of causality precise and to clear up the difficulties in which Mill and others are involved.\u201d p. 133 \u201cWe can now see what is meant by speaking of a necessary and sufficient condition of a certain type of occurrence within a certain field; and, assuming it to be a precedent condition, then it is what we call a “cause”. In other words, if there is any force in the line of argument so far pursued, a cause is always a cause within a field.\u201d [13] p. 134<\/p>\n

[vi]<\/span><\/a> \u201cwhen a cause is taken as a necessary and sufficient precedent condition of the occurrence of a<\/p>\n

phenomenon (its “effect”) in a certain field, then it follows that the effect is a necessary and sufficient subsequent condition of the occurrence ( or operation) of the cause in the field. So that, granting the temporal priority of the cause, there is no question of any logical priority; and while, if our causal beliefs are true, we can with certainty, given the cause, infer that the effect will occur, we can with equal force infer, given<\/p>\n

the effect, that the cause has occurred.\u201d [13] (p.135-6)<\/p>\n

[vii]<\/span><\/a> On some of time\u2019s properties, see Shoham [14], e.g.: \u201cPresumably, we will always want to assume that time is acyclic ( that is, that it is a partial order), although some would dispute even this assumption (for treatment of cyclic time see, e.g., [79,98]).\u201d & \u201cIn most of the literature in theoretical computer science one finds discrete time(the only exception of which I am aware is [5]).\u201d p. 35<\/p>\n

[viii]<\/span><\/a> \u201cThere are two parts to my claim that time does not exist. I start from the philosophical conviction that the only true things are complete possible configurations of the universe, unchanging Nows. Unchanging things do not travel in time from Now to Now. Material things, we included, are simply parts of Nows.\u201d\u00a0 [15] p. 49; \u201cthe important thing is to get away from the idea that time is something. Time does not exist. All that exists are things that change. What we call time is- in classical physics at least- simply a complex of rules that govern the change.\u201d p. 137;<\/p>\n

[ix]<\/span><\/a> Taylor says: \u201cTo estimate the uncertainty in the sum or difference, we had only to decide on their highest and lowest probable values. The highest and lowest probable values of x are x best<\/sub><\/strong> \u00b1 \u03b4x, and those of y<\/p>\n

are y best<\/sub><\/strong> \u00b1 \u03b4y.\u201d [16] p. 45 For products and ratios however, a different rule emerges: \u201cwhen quantities are multiplied or divided, the fractional uncertainties add\u201d p.\u00a0 53, where of a quantity x the fractional uncertainty is \u03b4x \/ |x best<\/sub><\/strong>|, where the absolute value is used to yield a positive value.<\/p>\n

[x]<\/span><\/a> \u201cReactivity is when the process of measurement induces change in the phenomenon. Such occurrences are most likely with attitude and value measures.\u201d [17] p. 211<\/p>\n

[xi]<\/span><\/a> A good intuition for the \u2018cohort effect\u2019 is the phenomenon of \u2018same age groups experiencing that age differently (or not); Garrison Keillor gives some clues: \u201cWhat is it like to be twenty-one or even thirty? And death \u2014 what\u2019s that like? What would you do if communists made you choose between renouncing God and drinking a pitcher of warm spit? What would it be like to put your arms around a girl? I\u2019d seen it done by older kids but never tried it myself. What would it feel like? Would we talk? Would I kiss her or should I wait for her to kiss me?\u201d A June morning, assessing the situation<\/a> In this vein, it is somewhat striking how little \u2018life experience\u2019 accumulates over evolutionary-range time intervals: same teenagers struggle with the same \u2018demons\u2019, despite hundreds of similar generations before them going through the same experiences: very little \u2018species-level knowledge\u2019 gets passed down (i.e. not much cohort effect\u2019 in this manner!).<\/p>\n

References<\/em><\/p>\n

    \n
  1. Chiang, C.L., The life table and its applications<\/em>, in Introduction to Stochastic Processes in Biostatistics<\/em>, C.L. Chiang, Editor. 1968, John Wiley & Sons: New York. p. 189-214.<\/li>\n
  2. Rothman, K.J., S. Greenland, and T.L. Lash, Modern epidemiology.<\/em> 2008.<\/li>\n
  3. Van Fraassen, B.C., An introduction to the philosophy of time and space.<\/em> 1970.<\/li>\n
  4. Arstila, V., et al., The illusions of time: Philosophical and psychological essays on timing and time perception<\/em>. 2019: Springer.<\/li>\n
  5. Pearl, J., Probabilistic reasoning in intelligent systems: networks of plausible inference <\/em>https:\/\/drive.google.com\/file\/d\/1gYtsPNIoFolgrveDF7kjLawU11yISmdO\/view?usp=sharing<\/em><\/a>. 1988: Morgan Kaufmann.<\/li>\n
  6. Simon, H.A., Spurious correlation: a causal interpretation.<\/em> Journal of the American statistical Association, 1954. 49<\/strong>(267): p. 467-479.<\/li>\n
  7. Simon, H.A., The meaning of causal ordering.<\/em> Qualitative and quantitative social research: Papers in Honor of Paul F. Lazarsfeld, 1979: p. 65-81.<\/li>\n
  8. Suppes, P., A Probabilistic Theory of Causality<\/em>. 1970, North Holland: Amsterdam.<\/li>\n
  9. Shoham, Y., Chronological ignorance: experiments in nonmonotonic temporal reasoning.<\/em> Artificial intelligence, 1988. 36<\/strong>(3): p. 279-331.<\/li>\n
  10. Iwasaki, Y. and H.A. Simon, Causality in device behavior.<\/em> Artificial intelligence, 1986. 29<\/strong>(1): p. 3-32.<\/li>\n
  11. De Kleer, J. and J.S. Brown, Theories of causal ordering.<\/em> Artificial intelligence, 1986. 29<\/strong>(1): p. 33-61.<\/li>\n
  12. Iwasaki, Y. and H.A. Simon, Theories of causal ordering: Reply to de Kleer and Brown<\/em>, in Readings in Qualitative Reasoning About Physical Systems<\/em>. 1990, Elsevier. p. 661-665.<\/li>\n
  13. Anderson, J., The problem of causality.<\/em> The Australasian Journal of Psychology and Philosophy, 1938. 16<\/strong>(2): p. 127-142.<\/li>\n
  14. Shoham, Y., Reasoning about change: time and causation from the standpoint of artificial intelligence<\/em>. 1987: Yale University.<\/li>\n
  15. Barbour, J., The end of time: The next revolution in physics<\/em>. 2001: Oxford University Press.<\/li>\n
  16. Taylor, J., Introduction to error analysis, the study of uncertainties in physical measurements<\/em>. 1997.<\/li>\n
  17. Bollen, K.A., Structural equations with latent variables<\/em>. 1989, New York: John Wiley and Sons.<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"

    *** Aging research invites some unique statistical modeling challenges: the natural course of development in the later years of life is generally marked by declines across the board. That\u2019s why many constructs describing such processes contain in their very cores this \u2018going down\u2019 direction of changes, like \u2018sarcopenia\u2019 (i.e. \u2018loss of sarcomeres\u2019, and hence of […]<\/p>\n","protected":false},"author":2514,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"_acf_changed":false,"footnotes":""},"acf":[],"publishpress_future_action":{"enabled":false,"date":"2026-04-12 05:03:38","action":"change-status","newStatus":"draft","terms":[],"taxonomy":""},"_links":{"self":[{"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/pages\/106"}],"collection":[{"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/users\/2514"}],"replies":[{"embeddable":true,"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/comments?post=106"}],"version-history":[{"count":4,"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/pages\/106\/revisions"}],"predecessor-version":[{"id":126,"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/pages\/106\/revisions\/126"}],"wp:attachment":[{"href":"https:\/\/health.uconn.edu\/causality\/wp-json\/wp\/v2\/media?parent=106"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}