Catnip immunity and alternatives

Estimation of catnip immunity rates by country with meta-analysis and surveys, and discussion of catnip alternatives. (statistics, psychology, meta-analysis)
created: 7 Nov 2015; modified: 14 Sep 2017; status: in progress; confidence: likely; importance: 6

Not all cats respond to the catnip stimulant; the rate of responders is generally estimated at ~70% of cats. A meta-analysis of catnip response experiments since the 1940s indicates the true value is ~62%. The low quality of studies and the reporting of their data makes examination of possible moderators like age, sex, and country difficult. Catnip responses have been recorded for a number of species both inside and outside the Felidae family; of them, there is evidence for a catnip response in the Felidae, and, more uncertainly, the Paradoxurinae, and Herpestinae.

Catnip (Nepeta cataria) is a plant which causes stimulated and excitable behavior in many domestic cats, which is fun to play with. It is the best known of at least a dozen plants with psychoactive effects on cats, and far more popular, cheaper, and easily purchased than alternatives like Tatarian honeysuckle, silvervine, or cat thyme. However, a large fraction of cats do not respond (non-response may be a genetic trait given Neil Bowman Todd 1962’s pedigree chart) but may respond to one of the alternatives. For example, of my family’s two cats, neither responds to catnip but one responds to valerian, one to honeysuckle, and both to silvervine.

This raises several questions:

  1. what fraction of cats, exactly, are catnip-immune?
  2. how often do the catnip alternatives work?
  3. does immunity to one alternative predict immunity to others?
  4. since some alternatives can be hard to find and expensive, what is the optimal order of alternatives a cat owner should try to find one that works?

Population frequency of catnip response

Literature review

Catnip is frequently discussed in the popular & secondary literature without citation, and after tracking down claims, the primary literature on catnip effects & response is relatively small1, with most research focused on the chemical synthesis of the active ingredients, botanical studies, or investigating the possible commercial applications as insecticide & insect repellent; the best literature review remains Tucker & Tucker 1987. Key papers:

  • McElvain et al 1942, The constituents of the volatile oil of catnip. II. The neutral components. Nepetalic anhydride:

    tested extracted nepetalactone and caryophyllene on 10 lions of unspecified but mixed genders/ages; the 7 adults all responded to the nepetalactone and not the caryophyllene (the 3 cubs responded to nothing).
  • Todd 1962, Inheritance of the catnip response in domestic cats

    14 responders, 12 non-responders in a Siamese breeding colony so 46% immunity rate in this sample. (8 male responders, 6 female responders, 2 male non-responders, 10 female non-responders.) Todd also surveyed cats in local animal shelters/hospitals, finding 26 of 84 sampled were non-responders or a ~31% immunity. Todd considers the genetic pattern most consistent with a fairly common genetic variant (by Hardy-Weinberg: p2+2pqp^2+2pq; p2+2pq=0.69;q2=0.31p^2 + 2pq = 0.69; q^2 = 0.31, then p=0.44;q=0.56p=0.44; q=0.56) which is autosomal dominant; this would imply that breeding cats for catnip response is highly feasible.
  • Todd 1963, The catnip response

    • In surveying the 26 breeding colony & 84 local cats, Todd found no large correlations with sex, breed type (Manx/Siamese/tabby/Agouti), white spotting, blue dilution, polydactyly, long hair, or castration. (32 male responders, 11 male non-responders, 26 female responders, 15 female non-responders.) Except the usual observation that young kittens rarely display a catnip response: of 39 under 12 weeks of age, 4 responded.
    • Cross-species results:

      1. Viverrids (pg42/73), unspecified gender/age/species:

        Table 3. Results of testing Viverrids with catnip. Figure in parentheses indicates number of individuals tested. + = positive, ?+ = possibly positive, ?- = probably negative, - = negative, I = indeterminate, animal would not or did not investigate leaves.
        Sub-families Genera N + ?+ ?- - I
        Viverrinae Genetta 3 0 0 0 2 1
        Viverrinae Viverra 1 0 0 1 0 0
        Viverrinae Civettictis 2 0 0 0 1 1
        Paradoxurinae Nandinia 5 0 0 3 2 0
        Paradoxurinae Paguma 2 0 0 1 0 1
        Paradoxurinae Arctictis 3 1 1 0 0 1
        Herpestinae Herpestes 1 0 0 0 1 0
        Herpestinae Atilax 3 0 1 1 1 0
        Herpestinae Ichneumia 1 0 0 0 0 1
        Cryptoproctinae Cryptoprocta 1 0 0 1 0 0
      2. Hyenas (Hyenidae): 0/3 responders of 2 males/1 female (pg41/72), unspecified age, genus, or species, presumably either spotted or striped hyenas.
      3. Felidae: (pg42/74). Todd’s table and results have been summarized as thus by Tucker & Tucker:

        Within the subfamily Pantherinae of the Felidae, Todd (1963) found the typical catnip response in 16 lions (Panthera leo) (14 positive responders, 2 negative responders), 23 tigers (Panthera tigris) (8 incomplete responders, 13 negative responders, 2 inconclusive responders), 18 leopards (Panthera pardus) (14 positive responders, 4 negative responders), 8 jaguars (Panthera onca) (7 positive responders, 1 negative responder), 4 snow leopards (Panthera uncia) (4 positive responders), and 1 clouded leopard (Neofelis [Panthera] nebulosa). Within the subfamily Acynonychinae of the Felidae, he found that 3 cheetahs (Acinonyx jubatus) did not respond to catnip. Within the subfamily Felinae of the Felidae, he found the typical catnip response in the 2 bobcats (Felis [Lynx] rufus) (1 positive responder, 1 negative responder), 1 European lynx (Felis [Lynx] lynx), 5 pumas (Felis [Puma] concolor) (2 positive responders, 2 negative responders, 1 inconclusive responder), 1 Asiatic golden cat (Felis [Profelis] temmincki), 5 ocelots (Felis [Leopardus] pardalis) (4 positive responders, 1 negative responder), and 6 margay cats (Felis [Leopardus] wiedii) (4 positive responders, 2 negative responders); no catnip response was observed in 2 servals (Felis [Leptailurus] serval), 1 swamp cat or jungle cat (Felis chaus), 1 Pallas’ cat (Felis [Octocolobus] manul), 1 leopard cat (Felis [Prionailurus] bengalensis), 1 African golden cat (Felis [Profelis] aurata), 2 fishing cats (Felis [Prionailurus] viverrina), 4 jaguarundis (Felis [Herpailurus/Puma] yagouaroundi), and 1 pampas cat (Felis [Lynchailurus] pajeros).

        The summary of Todd’s Felidae results is correct but omits that Todd also used a questionnaire to estimate 4 responders & 13 non-responders among the ocelots in addition to his personal testing of the 5 ocelots but Todd notes he has reservations about the validity of the data gathered by this method (possibly because the response frequency 4/17 is so different from his personal 4/5 response frequency) so that data is probably best excluded. Todd also mentions a report of a ligeress (female hybrid of a male lion & tigress) from de Bary of the Utah Hogle Zoological Garden who is a catnip non-responder. For the meta-analysis, I code the questionable positive/negative responses as confirmed responses. Sex is usually unspecified, but individual-level data is provided on the lions, tigers, & leopards. I excluded the pampas cat and additional questionnaire data in line with Todd’s notes that the data were highly unreliable.
  • Bates & Siegel 1963, Terpenoids. Cis-trans- and trans-cis- nepetalactones

    They chemically separated the two isomers; the extracted isomer I was mostly inactive when presented to 8 cats with just 1 more interested in I than II (though they caution there might still have been contaminating II in the I extract), but three were strongly attracted and two showed slight preference for isomer II (_trans-cis-_nepetalactone), so arguably 5 out of 8 were responders. (Breed and sexes unspecified.)
  • Konecny 1936, Behavioral Ecology of Feral House Cats in the Galapagos Islands, Ecuador: wild survey of cats; notes successful use of traps baited with tuna fish & catnip.
  • Palen & Goddard 1966, Catnip and oestrous behavior in the cat

    Reactions: 23 responders, 20 non-responders, so 47% immunity rate. (37 male, 28 female, mixed breeds: 6 male responders, 6 male non-responders, 9 castrated male responders, 5 castrated male non-responders, 5 female responders, 5 female non-responders, 3 spayed female responders, 4 spayed female non-responders.)
  • Hayashi 1968a, Pseudo-Affective Reflexes of Cats produced by Extracts from the Plant Actinidia polygama claims to have found no responses in an unspecified but probably >4 number of cats (as he used young and old cats of both sexes) when testing nepetalactone (catnip) and actinidine solutions, aside from two actinidine reactions
  • Hayashi 1968b, Motor reflexes of cats to Actinidia polygama (Japan) and to catnip (USA)

    A 1966 conference talk published in the 1968 proceedings, Hayashi 1968 is light on details. Tucker & Tucker summarize it as Hayashi (1968), who tested a wide range of animals (dogs, rabbits, mice, rats, guinea pigs, fowls, and cats) with powders of Actinidia polygama and N. cataria, found that the catnip response is induced in cats alone., which tells one about as much as the original report does:

    …actinidine (1) and catnip…have always been the source of much interest…When powder of these plants was presented to cats, they displayed a peculiar behavior…The reflex behavior is induced by the smell, not by taste and not via the circulation. The cat must be tamed by the experimenters and must be adult: male or female are quite the same…Once I tried with English cats in London, and with American cats in New York, each time taking advantage of visiting my friend’s laboratories, but English as well as American cats were rather cold…We tried experiments with dogs, rabbits, mice, rats, guinea pigs and also with fowls but they had no such reflexes to the plant powder…From these results, we presume that the reflex ability is restricted to cats and feline species in vertebrates, and its reflex centre would be situated in the subcortical, presumably limbic, structures…That the phenomenon is restricted to the cat family is said also in Japan.

    • Q [J.W. Johnson Jr]: This is a simple question that might be relevant to your point, Doctor. In Northern Virginia, where I live, there are stands or clumps of catnip. I’m not aware of house cats visiting the catnip stands while in the living state. Do you know whether this has been reported.
    • A [T. Hayashi]: No, I don’t think so. We have natural growths of bushes of actidinia polygama in several parts of Japan, but nobody noticed that these bushes attracted cats from the villages….In the central part of Japan we have many actidinia distributed from north to south. But I have not heard that cats gather in the stands of the plant. Maybe the drying of the plants or burning of them is the most effective.
    From this I gather that Hayashi must have tested at least 2 cats in the USA, 2 in the UK, and 2 in Japan (because he always uses the plural cats, and he compares the US/UK cats to Japanese cats). Hayashi 1968a implies >=4 cats were used but not their national distribution. To be conservative in lieu of more precise data, one would have to code the Hayashi data as 2 cats per country. He does mention that the USA/UK cats were rather cold, which implies an intermediate but existent response much less than the Japanese cats. (It is also intriguing given Sakurai et al 1988’s later possible implication that 16 of 16 cats in Japan reacted to catnip but not foreign breeds.) If we assume the n of each group of cats is 2, then the response rate must be 1/1/2 respectively, as otherwise Hayashi would either have described it as no response like the other species or as the same response as the Japanese cats. The other species are also pluralized, so at least 2 of each, and all responses must have all been 0 responders since they had no such reflexes and it is restricted to cats and feline species. Specific species (presumably the other animals are either domestic or lab species), sexes, breeds, and ages are not given.
  • Waller et al 1969, Feline Attractant, cis,trans-Nepetalactone: Metabolism in the Domestic Cat

    In a metabolic study, purified nepetalactone was force-fed to 6 cats in gel capsules; as expected due to the olfactory requirement (previously demonstrated by Todd with surgical manipulations of olfaction), none of the cats exhibited the catnip response and the result is irrelevant.
  • Hatch 1972, Effect of drugs on catnip (Nepeta cataria) induced pleasure behavior in cats

    Tested 17 cats initially; 14 responders. (Mixed breed, both genders; breakdowns of response not given.) Hatch’s main effort was towards using drug administrations to block the catnip effects and undermine the sexual hormone hypothesis.
  • Todd as quoted in R.F. Ewer 1973, The Carnivores (pg244):

    Differential aspects of the same scent on different species are illustrated in some unpublished work by Dr N.B. Todd, which he has kindly allowed me to quote. He tested the responses of various species to catnip and found that a number of viverrids, although clearly able to smell the catnip, showed little interest: for them it carried no message and had no particular significance. I have found the same to be true of Civettictis and Crossarchus [African dwarf mongoose]. In the Felidae, the cat-type sexual response was widespread but not universal. Amongst the large cats, lion, leopard, jaguar and snow leopard responded sexually but adult tigers did not. The behavior of immature tigers was interesting: they were not sexually excited but instead showed violent alarm and retreated promptly. In view of the propensity of the larger species of felids to kill each other’s young and of the fact that until quite recently the ranges of tiger and lion overlapped very considerably, both the absence of sexual response in the adult tiger and the fear generated in the young may be adaptive: the former may reflect a sexual isolating mechanism; the latter may be protective. Apart from illustrating differential responses to the same odour, these observations also demonstrate the psychological effects of odours. A piece of material impregnated with catnip is neither a mate nor an enemy but it can evoke sexual responses or flight: to do so it must act on the central nervous system so as to change the animal’s mood - in short, it makes him sexy or it makes him afraid.

    Todd’s results here refer to his PhD thesis, but I have not found any Ewer publication on catnip thus far.
  • Hill et al 1976, Species-characteristic Responses to Catnip by Undomesticated Felids; Tucker & Tucker summary (apparently based on counting the descriptions in Table 3 or Table 4):

    Hill et al. (1976) found that lions (5 positive responders, 6 partial responders, 1 negative responder) and jaguars (3 positive responders) are extremely sensitive to catnip, while tigers (5 negative responders), pumas (4 negative responders), leopards (4 partial responders, 4 negative responders), and bobcats (2 negative responders) gave little or no response. They also found that both males and females of the same species test alike, while reproductive-age adults are more sensitive than either aged or immature animals.

  • Harney et al 1978, Behavioral and toxicological studies of cyclopentanoid monoterpenes from Nepeta cataria

    Considers only injections of catnip oil into mice/rats.
  • Sakurai et al 1988, Both (4a_S_, 7_S_, 7a_R_) -(+)-Nepetalactone and Its Antipode Are Powerful Attractants for Cats

    Another investigation of which isomers/enantiomers of nepetalactone are active, the isolated versions were tested in 9 cats with vials of the liquid; 7 responded (while the two 6-month olds didn’t and Sakurai et al 1988 attributes the non-response to their being immature, 6 months sounds old enough for reactions to have developed). The cat breeds are specified as 4 Japanese, 3 Abyssinian, and 2 American short-hairs, but not sexes (although Sakurai used a mix, given their comments that the females showed more emotional behavior than the males…the females were quite reactive, while the males were not…both of them [isomers] were extremely attractive to mature cats, especially to females). Possibly contradicting Bates & Siegel 1963’s results where only 1 isomer worked, Sakurai finds both isomers work equally well.

    Confusingly, Sakurai also mentions a second experiment in impregnating filter paper with the isomers, noting that sixteen Japanese cats reacted to the 0.01mg dose; it’s unclear where these cats came from when only 9 cats (4 Japanese) were mentioned for the first experiment, and whether they were selected out of a larger group of cats or if it’s implied that it was 16 out of 16 responders - inasmuch as with catnip response rates ~70%, it would be highly improbable for all of a group of 16 cats to be catnip responders (0.716=0.3%). One possible explanation is that Japanese cats, being a historically isolated population (eg the Japanese Bobtail) after their introduction ~500AD from Korea, may have much higher population frequencies or even fixed the catnip mutation due to genetic drift or a founder effect2; on the other hand, if Japanese cats were almost all catnip responders, you would think someone would have noticed by now. My GS survey finds a ~28% immunity rate, which is not unusual at all, suggesting that Hayashi/Sakurai is some sort of sampling error or reporting effect.
  • DeLuca & Kranda 1992, Environmental enrichment in a large animal facility:

    TABLE 1. Number of animals, species-wide, that showed interest in various toys. Key: 0 = none, 1 = 25%, 2 = 50%, 3 = the majority of the animals. We define interest as the amount of toy destruction and/or movement that we noted in daily observations.

    • …balls: catnip: 3
    • balls: punch: 3
    • Cat-a-Comb: NA
    • Purrsuit: 3

    …8-12 cats…The cats’ favorite toys - like the 12 in. giant sheepskin mice (Petraport, Anaheim, CA) - contained catnip. We hung fresh catnip in stockinette bags which the cats quickly pulled down and ultimately batted into the water bowls. We gave them catnip-treated punchball toys (Petraport, Anaheim, CA- Fig. 7)-2 in. puffs mounted by a spring to a 6 in. x 6 in. base. The cats always played with these toys, and they lasted longer than the catnip bags. The cats also had Mr. Spats’ Cat-a-combs groomer (Tarel Seven Designs, Secaucus, NJ) mounted on the walls. They usually knocked these off [of] the wall and used them as play-things (they managed to open the compartment and dig out the catnip) rather than as grooming tools. Most of the cats spent time playing Purrsuit (Tarel Seven Designs, Secaucus, NJ)-every morning, we placed toys inside a maze and the cats chased them and tried, with a great deal of success, to get the smaller ones out. The cats had a preference for balls or bells with catnip in them, and for golf balls. They ignored the Squish balls (Ethical Inc., Newark, NJ).

    So based on the reported data, we can guess that at least 5-7 cats responded to catnip of the 8-12 sample, for a best guess of 6 responders out of 10 cats. (Mixed breeds, unspecified sex.)
  • Eason et al 1992, Toxic Bait and Baiting Strategies for Feral Cats/Clapperton et al 1992, Experimental Eradication of Feral Cats (Felis Catus) from Matakohe (Limestone) Island, Whangarei Harbour: New Zealand, a mix of wild survey and experiment, employing a variety of baits & scents including catnip. Catnip performed well among the NZ feral cats: for example, 6 cats observed at dusk spent a mean ~200s investigating a catnip odor as opposed to <50s or ~0s for 3 urine scents (Figure 1), and counts of visits to catnip odor stations were far higher than for urine, fish oil, or water controls (Table 1: 25 vs 13/12/11). Catnip/catmint only performed poorly when used as a food flavoring in a commercial cat food (Figure 3), but then performing well alone & combined with L-alanine in bait (Figure 5/6). They then deployed their catnip & other baits to poison ~5 cats on Matakohe Island; the catnip bait did not lead to noticeably more bait consumption, but from the description, catnip response might not have been noticeable or just sampling error.
  • Clapperton et al 1994, Development and Testing of Attractants for Feral Cats, Felis catus L.

    Wild survey and experiment in New Zealand. Catnip & silvervine were tested as cat lures for trapping; simultaneous testing over multiple environments showed that catnip, silvervine, and urine all garnered substantial attention from cats. Clapperton notes that 4/4 domestics and 8/20 feral cats responded (pg7). (Sex not specified, but breeds were clearly mixed as feral cats are never single breeds.)
  • Harrison 1997, Chemical attractants for Central American felids: experimental testing of jaguars, jaguarundi, little spotted cats, margays, ocelots, & pumas of catnip and other lures; Harrison reports the data in terms of time spent investigating and behavior scores (the number of observed behaviors such as sniffing/vocalizing/rubbing/rolling), and not in terms of individual responders. The reported data suggests some level of response in jaguars/jaguarundi/ocelots.
  • Edwards et al 1997, Field evaluation of olfactory lures for feral cats (Felis catus L.) in central Australia: wild survey in Australia; 15 kinds of lures were tested, pitting a catnip lure (#12) against various seafood, blood-bone mix, male/female cat urine/anal gland secretions; the 3 best olfactory lures were sun-dried prawns (#8) & male (#13)/female (#14) urine-secretions, but not catnip - catnip received 2 cat visits while the best 3 received 17/14/7 respectively. Edwards notes their surprise at the inefficacy of the catnip compared to the urine-secretions, as Clapperton et al 1994 had found catnip much superior in their NZ experiment and ask, do cats from different areas respond differently to different olfactory stimuli?
  • McDaniel et al 2000, Efficacy of lures and hair snares to detect lynx

    Used catnip and other commercial products in scent stations to look for activity of wild lynx. Todd 1963 had already found responses in lynx/puma/bobcats, so unsurprisingly McDaniel does too, but with wild lynx, it is impossible to know how many total lynx were exposed and how many reacted.
  • Molsher 2001, Trapping and demographics of feral cats (Felis catus) in central New South Wales: Australian live trapping study using varied baits; a bait combination of catnip/tuna oil/synthetic fermented eggs (SFE); while a small sample, the success in terms of cats trapped per 100 trap-nights varied from 0.5 (fish) to 3.8 (PUSSON baits + aluminum toys on string), with the catnip mix at 1.1, suggesting little or no advantage over other baits like rabbit.
  • Short et al 2002, Control of feral cats for nature conservation. III. Trapping: Australian trapping study, varying bait and timing. Fresh catnip bait ranked roughly median, below rabbit/mice/2 commercial mixes; catnip worked well in the first trapping and then poorly afterwards, the authors speculating that the decline is perhaps due to removing the pool of susceptible individuals. (Total catching: 8 cats out of 118; in a head to head test of 2 traps, one catnip and a commercial mix, the catnip trap caught 8 vs 7.)
  • Wells & Egli 2004, The influence of olfactory enrichment on the behaviour of captive black-footed cats, Felis nigripes:

    6 black-footed cats (Felis nigripes), catnip did interest them and cause increases in activity, but paper doesn’t break down by cat. Not useful unless want to contact authors for individual-level data.
  • Weaver et al 2005, Use of scented hair snares to detect ocelots: tested 32 ocelots available in 9 facilities, as well as a wild survey; 27 of the 32 responded to their bait, which was a combination of a mix Weaver’s Cat Call and dried catnip; Weaver reports that the combination was earlier established in bobcats & lynx to be much more effective than catnip alone, so 84% represents an overestimate of catnip’s effect on ocelots.
  • Ellis 2007, Sensory enrichment for cats (Felis silvestris catus) housed in an animal rescue shelter; apparently republished as Ellis & Wells 2010, The influence of olfactory stimulation on the behaviour of cats housed in a rescue shelter

    Shelter animals were given catnip-infused clothes to play with; Ellis notes that the catnip toys were played with more than other scents on average in the catnip group, but made no effort to ascertain how many were catnip responders.
  • Massoco et al 1995, Behavioral effects of acute and long-term administration of catnip (Nepeta cataria) in mice; Bernardi et al 2010, Antidepressant-like effects of an apolar extract and chow enriched with Nepeta cataria (catnip) in mice

    There is apparently a vein of studies trying catnip in humans for antidepressant effects (rather than the more traditional painkiller and psychedelic effects), leading to this experiment in chronic feeding catnip to mice (ironic as that might sound), finding one antidepressant-like effect. These can’t be considered a catnip response, though.
  • Schmidt & Kowalczyk 2006, Using Scent-Marking Stations to Collect Hair Samples to Monitor Eurasian Lynx Populations / Davoli et al 2013, Hair snaring and molecular genetic identification for reconstructing the spatial structure of Eurasian lynx populations: wild survey using catnip oil and beaver oil; 29 unique individuals genotyped
  • Downey et al 2007, Hair Snares for Noninvasive Sampling of Felids in North America: Do Gray Foxes Affect Success?: wild sampling in Mexico; catnip oil did not attract any of the target margay wild cats, but did attract many gray foxes & 14 domestic cats.
  • Long et al 2007, Comparing Scat Detection Dogs, Cameras, and Hair Snares for Surveying Carnivores: wild survey, used dried catnip; no detection of bobcats
  • Castro-Arellano et al 2008, Hair-Trap Efficacy for Detecting Mammalian Carnivores in the Tropics: wild survey in Mexico, comparing perfume & catnip oil, finding catnip oil inferior
  • Resende et al 2011, Influence of Cinnamon and Catnip on the Stereotypical Pacing of Oncilla Cats (Leopardus tigrinus) in Captivity

    The 8 oncilla cats’ activity were measured over several days after each dose was introduced into their enclosures. While the doses were small (1g), the cinnamon produced statistically-significant overall average difference while catnip did not, suggesting none of the oncilla cats responded to the catnip. Nevertheless, like Wells & Egli 2004, cannot be meta-analyzed.
  • Comer et al 2011, Bobcats Do Not Exhibit Rub Response Despite Presence at Hair Collection Stations: wild survey using catnip oil; no near-zero rubbing
  • Matthew 2012, A Comparison of Noninvasive Survey Methods for Monitoring Mesocarnivore Populations in Kentucky: wild survey; did not break down detection rates by lure type
  • Hanke & Dickman 2013, Sniffing out the stakes: hair-snares for wild cats in arid environments: Australian wild survey; catnip & valerian proved ineffective in attracting cats to the stakes, and they suggest that Australian wildcats (apparently descended from domesticated cats) may simply be catnip immune, noting their results parallel Molsher 2001 & Short et al 2002. (This may be connected to Australia’s import regulations which deter bringing new cats into the country: a number of hybrid breeds are banned as well as origin countries; and legal cats must be: permitted, microchipped, rabies-vaccinated at least 180 days in advance & tested to confirm immunity, not be too pregnant, treated for internal parasites, treated for external parasites, clinically examined by a Australian-government-approved veterinarian, and quarantined for at least 10 days.)
  • Portella et al 2013, Assessing the efficacy of hair snares as a method for noninvasive sampling of Neotropical felids (Table 1): wild survey and experiment of cinnamon/vanilla/catnip scent lures on 5 jaguars (Panthera onca)/10 ocelots (Leopardus pardalis)/6 oncilla cats (Leopardus tigrinus)/7 margay cats (Leopardus wiedii)/6 pumas (Puma concolor)/5 jaguarundis (Puma yagouaroundi). The wild survey yielded no results, while the captive experiment indicated that the puma did not interact with any lures and at least some of the ocelot/margay/oncilla did, with the strongest effect being vanilla.

    As Portella et al 2013 measures behavior in length of time interacting with the scent lures and reports only species-level differences, it is impossible to say what fraction of the cats were responders, other than the total lack of response of the 6 pumas to any scents showing that 0/6 of them responded.
  • Poddar-Sarkar & Brahmachary 2014, Chapter 15: Pheromones of Tiger and Other Big Cats note an unreported attempt at measuring catnip response rates in wild African lions:

    Certain persons brought catnip from England to George Adamson’s lion camp in Kora, Kenya, and tried to study the effect of this plant on the African lion but no conclusive results were obtained (Adamson 1988, personal communication).

  • Patkó et al 2015, Sneaky felids, smelly scents: a small scale survey for attracting cat: wild survey in Hungary, comparing catnip with salmon oil/valerian/commercial-scent; 1 stone marten & 3 cats rubbed the catnip, 4 valerian, none otherwise.
  • Patkó et al 2016, More Hair than Wit: A Review on Carnivore Related Hair Collecting Methods: literature review on wild surveying; no particular conclusions drawn on catnip.
  • Crowley & Hodder 2017, An assessment of the efficacy of rub stations for detection and abundance surveys of Canada lynx (Lynx canadensis): a wild survey; the lynx did use both the beaver castor & catnip oil rub stations but, like McDaniel et al 2000, the method does not permit any quantification of response rates in the Canada lynx although the authors infer it was <100% since the catnip oil rub stations did not work as well as the standard camera traps.
  • Shreve et al 2017, Social interaction, food, scent or toys? A formal assessment of domestic pet and shelter cat (Felis silvestris catus) preferences: experimental test of 25 pet & 25 animal shelter cats’ preferences by offering the stimuli simultaneously and measuring proportion of time on each to extract a choice; for example, in the olfactory category, of the 38 cats, 6 chose the gerbil, 6 the cloth rubbed on another cat’s scent glands (conspecific), and 22 the catnip. However, while 2238=0.58\frac{22}{38}=0.58 may look like a plausible estimate of catnip response in this sample, that would be making some strong assumptions, as Shreve et al do not mention trying to classify catnip response, and the choice is ambiguous: 22 of 38 cats finding catnip the most interesting smell does not mean 22 had a catnip response - as any of the 22 might not have had a response but simply found the catnip a more pleasing or novel smell, and vice versa for the other.
  • Bol et al 2017, Responsiveness of cats (Felidae) to silver vine (Actinidia polygama), Tatarian honeysuckle (Lonicera tatarica), valerian (Valeriana officinalis) and catnip (Nepeta cataria): one of the most thorough studies, using both a large number of cats (n=100) and the 4 major cat stimulants, Bol et al 2017 is an impressive experiment. It’s worth highlighting that the 4 stimulants were offered to the same set of cats, allowing measurement of intercorrelations, cats were exposed at least twice with attention given to a low-stress administration (reducing measurement error, as emphasized by Villani 2011), gas chromatography was used for a chemical analysis, the sample size is one of the largest ever (exceeded only by Villani 2011/Lyons 2013 and my surveys), the partnership with Big Cat Rescue extends the results to several other interesting species, and the full dataset is included with the paper.

TODO:

Potentially relevant but currently unavailable:

  • Hart, B.L., 1977. Olfaction and feline behavior. Feline Practice 7(5): 8-10 (requested)
  • Harney J.W., Leary JD, Barofsky IB. Behavioral activity of catnip and its constituents: nepetalic acid and nepetalactone, Federation Proceedings 1974; 33: 481 (requested)
  • Lyons 2013, Genome-Wide Association Study for Catnip Response in Domestic Cats: unpublished research, but abstract of results; TODO: email Leslie Lyons and ask if published anywhere or if the dataset can be made public; the catnip response rate would be informative, there may be other genetic analyses, and would boost any future GWAS:

    Results: No Significant Genetic Region Identified for Catnip Response in Cats. About 50% of cats respond to catnip. Funded by the Cat Health Network, researchers from the University of California–Davis tested 192 shelter cats for catnip response in controlled settings. DNA was collected from cats responding to catnip and compared to DNA of nonresponding cats. Genetic analysis of these samples did not reveal a causative gene associated with catnip response. Identification of genes responsible for catnip response may provide clues to the mechanisms involved in olfactory responses to drugs and chemicals in cats.

Overall, study quality is low and at high risk of bias. Key variables like sex/age/breed is almost always never reported and sometimes even n or species is not reported; placebo controls are not used; the experimenters are never blind to the substance being used; important studies are not available in English; the catnip used is not standardized; and the animals are not necessarily familiarized & comfortable with the experimenter despite the need for them to be relaxed and willing to play in order to gauge the presence or absence of a response. Probably any estimate is something of a lower bound as most of these biases would tend to mask a catnip response (like using old catnip or an animal feeling stressed), and it will be difficult to prove or disprove any sex or breed effects.

Data

catnip <- read.csv(stdin(), header=TRUE)
Family,Genus,Species,Name,Study, Year, Responders, N, Country, Sex, Breed
Felidae,Panthera,leo,African lion,McElvain et al 1942,1942,7,7,USA,mixed,NA
Viverrinae,Civettictis,NA,NA,Todd 1962,1962,0,3,USA,mixed,NA
Viverrinae,Viverra,NA,NA,Todd 1962,1962,0,1,USA,mixed,NA
Viverrinae,Civettictis,NA,NA,Todd 1962,1962,0,2,USA,mixed,NA
Paradoxurinae,Nandinia,NA,NA,Todd 1962,1962,0,5,USA,mixed,NA
Paradoxurinae,Paguma,NA,NA,Todd 1962,1962,0,2,USA,mixed,NA
Paradoxurinae,Arctictis,NA,NA,Todd 1962,1962,2,3,USA,mixed,NA
Herpestinae,Herpestes,NA,NA,Todd 1962,1962,0,1,USA,mixed,NA
Herpestinae,Atilax,NA,NA,Todd 1962,1962,1,3,USA,mixed,NA
Herpestinae,Ichneumia,NA,NA,Todd 1962,1962,0,1,USA,mixed,NA
Cryptoproctinae,Cryptoprocta,NA,NA,Todd 1962,1962,0,1,USA,mixed,NA
Hyaenidae,NA,NA,hyena,Todd 1962,1962,0,1,USA,F,NA
Hyaenidae,NA,NA,hyena,Todd 1962,1962,0,2,USA,M,NA
Felidae,Panthera,leo,African lion,Todd 1962,1962,6,6,USA,M,NA
Felidae,Panthera,leo,African lion,Todd 1962,1962,8,10,USA,F,NA
Felidae,Panthera,tigris,tiger,Todd 1962,1962,1,9,USA,M,NA
Felidae,Panthera,tigris,tiger,Todd 1962,1962,7,13,USA,F,NA
Felidae,Panthera,leo/tigris,ligeress,Todd 1962,1962,0,1,USA,F,NA
Felidae,Panthera,pardus,leopard,Todd 1962,1962,6,8,USA,M,NA
Felidae,Panthera,pardus,leopard,Todd 1962,1962,8,10,USA,F,NA
Felidae,Panthera,onca,jaguar,Todd 1962,1962,4,5,USA,M,NA
Felidae,Panthera,onca,jaguar,Todd 1962,1962,3,3,USA,F,NA
Felidae,Panthera,uncia,snow leopard,Todd 1962,1962,2,2,USA,M,NA
Felidae,Panthera,uncia,snow leopard,Todd 1962,1962,2,2,USA,F,NA
Felidae,Neofelis,nebulosa,clouded leopard,Todd 1962,1962,1,2,USA,NA,NA
Felidae,Acinonyx,jubatus,cheetah,Todd 1962,1962,0,3,USA,NA,NA
Felidae,Felis,rufus,bobcat,Todd 1962,1962,1,2,USA,NA,NA
Felidae,Felis,lynx,European lynx,Todd 1962,1962,1,1,USA,NA,NA
Felidae,Felis,concolor,puma,Todd 1962,1962,2,3,USA,M,NA
Felidae,Felis,concolor,puma,Todd 1962,1962,1,2,USA,F,NA
Felidae,Felis,temmincki,African golden cat,Todd 1962,1962,1,1,USA,NA,NA
Felidae,Felis,aurata,Asian golden cat,Todd 1962,1962,0,1,USA,NA,NA
Felidae,Felis,pardalis,ocelot,Todd 1962,1962,4,5,USA,NA,NA
Felidae,Felis,wiedii,margay,Todd 1962,1962,4,6,USA,NA,NA
Felidae,Felis,serval,serval,Todd 1962,1962,0,2,USA,NA,NA
Felidae,Felis,chaus,swamp cat,Todd 1962,1962,0,1,USA,NA,NA
Felidae,Felis,manul,Pallas' cat,Todd 1962,1962,0,1,USA,NA,NA
Felidae,Felis,bengalensis,leopard cat,Todd 1962,1962,0,1,USA,NA,NA
Felidae,Felis,viverrina,fishing cat,Todd 1962,1962,0,2,USA,NA,NA
Felidae,Felis,yagouaroundi,jaguarundi,Todd 1962,1962,0,4,USA,NA,NA
Felidae,Felis,pajeros,pampas cat,Todd 1962,1962,0,1,USA,NA,NA
Felidae,Felis,catus,cat,Todd 1962,1962,8,10,USA,M,Siamese
Felidae,Felis,catus,cat,Todd 1962,1962,6,16,USA,F,Siamese
Felidae,Felis,catus,cat,Todd 1963,1963,32,43,USA,M,mixed
Felidae,Felis,catus,cat,Todd 1963,1963,26,41,USA,F,mixed
Felidae,Felis,catus,cat,Bates & Siegel 1963,1963,5,8,USA,NA,NA
Felidae,Felis,catus,cat,Palen & Goddard 1966,1966,15,26,USA,M,mixed
Felidae,Felis,catus,cat,Palen & Goddard 1966,1966,8,17,USA,F,mixed
Felidae,Felis,catus,cat,Hayashi 1968,1966,1,2,USA,NA,NA
Felidae,Felis,catus,cat,Hayashi 1968,1966,1,2,UK,NA,NA
Felidae,Felis,catus,cat,Hayashi 1968,1966,2,2,Japan,NA,NA
Canidae,Canis,lupus,dog,Hayashi 1968,1966,0,2,Japan,NA,NA
Leporidae,Oryctolagus,cuniculus,rabbit,Hayashi 1968,1966,0,2,Japan,NA,NA
Muridae,Mus,musculus,mouse,Hayashi 1968,1966,0,2,Japan,NA,NA
Muridae,Rattus,norvegicus,rat,Hayashi 1968,1966,0,2,Japan,NA,NA
Caviidae,Cavia,porcellus,guinea pig,Hayashi 1968,1966,0,2,Japan,NA,NA
Phasianidae,NA,NA,fowl,Hayashi 1968,1966,0,2,Japan,NA,NA
Felidae,Felis,catus,cat,Hatch 1972,1972,14,17,USA,mixed,mixed
Felidae,Panthera,leo,African lion,Hill et al 1976,1976,3,4,USA,M,NA
Felidae,Panthera,leo,African lion,Hill et al 1976,1976,5,5,USA,F,NA
Felidae,Panthera,onca,jaguar,Hill et al 1976,1976,1,1,USA,M,NA
Felidae,Panthera,onca,jaguar,Hill et al 1976,1976,2,2,USA,F,NA
Felidae,Panthera,pardus,leopard,Hill et al 1976,1976,2,4,USA,M,NA
Felidae,Panthera,pardus,leopard,Hill et al 1976,1976,2,4,USA,F,NA
Felidae,Panthera,tigrus,tiger,Hill et al 1976,1976,0,1,USA,M,NA
Felidae,Panthera,tigrus,tiger,Hill et al 1976,1976,0,4,USA,F,NA
Felidae,Felis,rufus,bobcat,Hill et al 1976,1976,0,1,USA,M,NA
Felidae,Felis,rufus,bobcat,Hill et al 1976,1976,0,1,USA,F,NA
Felidae,Puma,concolor,cougar,Hill et al 1976,1976,0,2,USA,M,NA
Felidae,Puma,concolor,cougar,Hill et al 1976,1976,0,2,USA,F,NA
Felidae,Felis,catus,cat,Sakurai et al 1988,1988,4,4,Japan,mixed,Japanese
Felidae,Felis,catus,cat,Sakurai et al 1988,1988,3,3,Japan,mixed,Abyssinian
Felidae,Felis,catus,cat,Sakurai et al 1988,1988,2,2,Japan,mixed,American short-hair
Felidae,Felis,catus,cat,Sakurai et al 1988,1988,16,16,Japan,NA,Japanese
Felidae,Felis,catus,cat,DeLuca & Kranda 1992,1992,6,10,USA,NA,mixed
Felidae,Felis,catus,cat,Clapperton et al 1994,1994,12,24,USA,NA,mixed
Felidae,Felis,concolor,puma,Portella et al 2013,2013,0,6,USA,NA,NA


catnip$Sex <- ordered(catnip$Sex, levels=c("F", "mixed", "M"))
catnip$Country <- relevel(catnip$Country, "USA")

Meta-analysis

Cats’ catnip response rate

A random-effects meta-analysis on the proportion of catnip response in domestic cats using metafor:

cat <- subset(catnip, Name=="cat")
sum(cat$Responders) / sum(cat$N)
# [1] 0.6625514403

library(metafor)
r <- rma(xi=Responders, ni=N, measure="PR", slab=Study, data=cat); summary(r)
# Random-Effects Model (k = 17; tau^2 estimator: REML)
#
#   logLik  deviance       AIC       BIC      AICc
#   3.4814   -6.9627   -2.9627   -1.4176   -2.0397
#
# tau^2 (estimated amount of total heterogeneity): 0.0218 (SE = 0.0133)
# tau (square root of estimated tau^2 value):      0.1476
# I^2 (total heterogeneity / total variability):   66.56%
# H^2 (total variability / sampling variability):  2.99
#
# Test for Heterogeneity:
# Q(df = 16) = 60.1049, p-val < .0001
#
# Model Results:
#
# estimate      se     zval    pval   ci.lb   ci.ub
#   0.6960  0.0485  14.3376  <.0001  0.6008  0.7911

png(file="~/wiki/images/catnip-forest.png", width = 590, height = 610)
forest(r, xlim=c(0,1))
invisible(dev.off())

## some issues in the funnel plot of too-extreme values:
funnel(r)
trimfill(r)
# Estimated number of missing studies on the right side: 5 (SE = 2.7611)
#
# Random-Effects Model (k = 22; tau^2 estimator: REML)
#
# tau^2 (estimated amount of total heterogeneity): 0.0358 (SE = 0.0170)
# tau (square root of estimated tau^2 value):      0.1892
# I^2 (total heterogeneity / total variability):   74.75%
# H^2 (total variability / sampling variability):  3.96
#
# Test for Heterogeneity:
# Q(df = 21) = 82.4496, p-val < .0001
#
# Model Results:
#
# estimate      se     zval    pval   ci.lb   ci.ub
#   0.7694  0.0514  14.9565  <.0001  0.6686  0.8703

## Examine all moderators:
rall <- rma(xi=Responders, ni=N, measure="PR", slab=Study,
            mods= ~ Year + Country + Sex + Breed, data=cat)
summary(rall)
# Mixed-Effects Model (k = 10; tau^2 estimator: REML)
#
#   logLik  deviance       AIC       BIC      AICc
#   1.4148   -2.8297   15.1703    3.4087  195.1703
#
# tau^2 (estimated amount of residual heterogeneity):     0.0014 (SE = 0.0111)
# tau (square root of estimated tau^2 value):             0.0374
# I^2 (residual heterogeneity / unaccounted variability): 12.35%
# H^2 (unaccounted variability / sampling variability):   1.14
# R^2 (amount of heterogeneity accounted for):            90.59%
#
# Test for Residual Heterogeneity:
# QE(df = 2) = 2.6463, p-val = 0.2663
#
# Test of Moderators (coefficient(s) 2:8):
# QM(df = 7) = 15.3150, p-val = 0.0322
#
# Model Results:
#
#                           estimate       se     zval    pval     ci.lb     ci.ub
# intrcpt                   111.3234  64.3759   1.7293  0.0838  -14.8510  237.4979
# Year                       -0.0563   0.0328  -1.7192  0.0856   -0.1206    0.0079
# CountryJapan                1.1128   0.6297   1.7673  0.0772   -0.1213    2.3469
# Sex.L                       0.1231   0.0588   2.0950  0.0362    0.0079    0.2383
# Sex.Q                      -0.5262   0.2325  -2.2633  0.0236   -0.9818   -0.0705
# BreedAmerican short-hair   -0.0417   0.2765  -0.1507  0.8802   -0.5836    0.5002
# BreedJapanese               0.0250   0.2194   0.1139  0.9093   -0.4051    0.4551
# Breedmixed                  0.1600   0.1215   1.3170  0.1878   -0.0781    0.3981

r_country <- rma(xi=Responders, ni=N, measure="PR", slab=Study,
            mods= ~ Country, data=cat)
summary(r_country)
# Mixed-Effects Model (k = 17; tau^2 estimator: REML)
#
#   logLik  deviance       AIC       BIC      AICc
#   7.3649  -14.7298   -6.7298   -4.1736   -2.2854
#
# tau^2 (estimated amount of residual heterogeneity):     0.0063 (SE = 0.0071)
# tau (square root of estimated tau^2 value):             0.0795
# I^2 (residual heterogeneity / unaccounted variability): 33.42%
# H^2 (unaccounted variability / sampling variability):   1.50
# R^2 (amount of heterogeneity accounted for):            70.99%
#
# Test for Residual Heterogeneity:
# QE(df = 14) = 18.4673, p-val = 0.1863
#
# Test of Moderators (coefficient(s) 2:3):
# QM(df = 2) = 14.9306, p-val = 0.0006
#
# Model Results:
#
#               estimate      se     zval    pval    ci.lb   ci.ub
# intrcpt         0.6275  0.0417  15.0596  <.0001   0.5459  0.7092
# CountryJapan    0.2962  0.0775   3.8194  0.0001   0.1442  0.4482
# CountryUK      -0.1275  0.3648  -0.3496  0.7266  -0.8425  0.5874

power.prop.test(power=0.8, p1=0.62, p2=0.72+0.296)
#      Two-sample comparison of proportions power calculation
#
#               n = 13.66111205
#              p1 = 0.62
#              p2 = 1.016
#       sig.level = 0.05
#           power = 0.8
#     alternative = two.sided
#
# NOTE: n is number in *each* group

## moderately informative Bayesian meta-analysis
## (bayesmeta doesn't support moderators/meta-regression, so estimate just mean/SD):
library(bayesmeta)
b <- bayesmeta(y = escalc(xi=Responders, ni=N, measure="PR", data=cat),
               mu.prior.mean=0.5, mu.prior.sd=0.1); summary(b)
# ...marginal posterior summary:
#                     tau            mu        theta
# mode      0.15297890051 0.65753755334 0.6631979195
# median    0.16491992203 0.65361698541 0.6555978530
# mean      0.17168291630 0.65153464683 0.6515346468
# sd        0.04879271665 0.04995977983 0.1855800216
# 95% lower 0.08706766910 0.55130575642 0.2745600927
# 95% upper 0.26943049971 0.74754426677 1.0188759488
#
# (quoted intervals are shortest credibility intervals.)
#
# relative heterogeneity I^2 (posterior median): 0.7129601315
Meta-analytic forest plot of catnip response rates in experimental studies
Meta-analytic forest plot of catnip response rates in experimental studies

The 2/3 rule of thumb proves to be about right, with a meta-analytic summary of ~63% catnip response rates.

The heterogeneity of the results indicates that somethings are different from study to study or that biases like publication bias are at play; including the sex and breed covariates doesn’t help much because of the missing data, but there’s interesting trends of decreases with a study being conducted later in time and inside the USA, in line with my speculation about the Japanese studies finding catnip sensitivity to be almost universal. (The suspicious Sakurai 16/16 datapoint doesn’t show up in the full model because it has no sex information and gets dropped, so the Japanese trend there is being driven by the other Sakurai datapoints; the Hayashi results further support the Japanese anomaly.) The decrease with time is odd, but it’s much more likely that it’s reflecting differences in experimental procedure or breeds or countries than indicating that catnip response rates are being heavily selected against.3 Overall, the data quality is low as many authors (especially the chemists) did not report individual response rates or key details of the cats used such as their age, breed, or sex, which is particularly unfortunate as all the samples are small enough that the original data could’ve been easily included as short tables. This is ironic because while some authors claim that sex/breed doesn’t matter at all, others claim that females react much more strongly (or was it males?), and yet most don’t report the data down by those variables, so their results can’t be pooled for an answer.

Cross-species catnip response rates

Catnip has been tested in a number of species. It would be interesting to include this data for several reasons: it can help the search for the genetic basis by comparing catnip-sensitive species with non-sensitive looking for genetic variants peculiar to the former, or if there is no apparent cluster of closely-related species which uniquely have catnip responses, it can test theories about what local environments might cause catnip sensitivity; it can potentially help sharpen estimates of sex/age/country/experimenter effects by borrowing strength; zookeepers can enrich their animals’ lives with catnip if they know which species respond; and since the data is already there in the papers, we might as well include it.

To examine cross-species trends, we would like to fit a multilevel model using family/genus/species, since one would expect the catnip response rate to be more similar between, say, African lions & tigers than between snow leopards & hyenas. Since each species might have different gender effects and population bottlenecks, we would ideally nest in each species the sex/country/breed/year covariates. We would also like independent random-effects for each study to allow for the heterogeneity we observed before (which could be independent of the family/genus/species hierarchy, in which case it would model things like different experimental procedures - eg catnip leaves vs extract, or the experimenter being good or bad with animals - or it could be species-specific too). Unfortunately, such a full model would require estimating a huge number of parameters (with 24 species, 14 genuses, & 6 families, 10 studies, and 4 variables for each species, that’s easily 270 parameters which must be fit) and there is far too little data (and the available data is compromised by the frequent missingness of species and sex), so we have to settle for something more modest, focusing on just the cross-species aspect.

## full model is unidentifiable due to too few data points and missingness:
# b <- bglmer(cbind(Responders, N) ~ (Sex+Country+Breed+Year|Species/Genus/Family) + (1|Study), family="binomial", data=catnip)

library(metafor)
# levels(catnip$Name) <- c(levels(catnip$Name), as.character(catnip[is.na(catnip$Name),]$Genus))
# catnip[is.na(catnip$Name),]$Name <- as.character(catnip[is.na(catnip$Name),]$Genus)
## sort by taxonomy:
catnip <- catnip[with(catnip, order(Family, Genus, Species)),]
catnip$Label <- with(catnip, Map(function(f,g,s, n) { paste0(n, " (", paste(f, g, s), ")") }, Family, Genus, Species, Name))

cross <- rma(xi=Responders, ni=N, measure="PR", slab=Label,
            data=catnip)
summary(cross)

png(file="~/wiki/images/catnip-forest-crossspecies.png", width = 580, height = 1000)
forest(cross)
invisible(dev.off())

library(blme)
b <- bglmer(cbind(Responders, N) ~ (1|Species/Genus/Family) + (1|Study) + Year, family="binomial", data=catnip)
summary(b); ranef(b)

Survey

Google Surveys: Probability Population Sample

Google Consumer Surveys/Google Surveys (GS) is a online survey service offered by Google to the general population; it is powered by media publishers where their article paywalls are replaced by a short typically single-question survey of a variety of types which the reader must answer to see the desired content. Combining metadata with Google’s advertising profiles, Google attempts to weight selected readers for the surveywall such that they form an unbiased random probability sample of a specified population, such as the general population. Each GS response costs $0.10-$3 depending on the length of the survey & difficulty of recruiting responses (eg a country-wide all-ages all-gender probability sample survey with one question would cost $0.10 per response, so $10 would yield 100 answers).

As of August 2017, GS allows targeting:

  1. by age bracket:

    • 18-24
    • 25-34
    • 35-44
    • 44-54
    • 55-64
    • 65+
  2. gender: women/men
  3. country (and by sub-country, specific regions depending on country):

    • United States (English)
    • Canada (English)
    • United Kingdom (English)
    • Germany (German)
    • Mexico (Spanish)
    • Japan (Japanese)
    • Australia (English)
    • Brazil (Portuguese)
    • France (French)
    • Spain (Spanish)
    • for Android smartphone readers only, Italy (Italian) and the Netherlands (Dutch but not English)
  4. by predefined groups, audience panels (costing more, >=$0.30); eg for the United States, available audience panels include

    • Hispanics/Latinos
    • Online dating users (Sites and apps)
    • Small/Medium Business Owners and Managers
    • Mobile social media users (Facebook, Twitter, Google+)
    • Streaming video subscription users (Netflix, Amazon, Hulu Plus, Google Play)
    • Students
  5. by user-defined screening question which provides a conditional question, and one is charged only for the users who answer positively & proceed to the main survey question

For catnip surveying, GS has advantages and disadvantages. The primary advantages are that a well-powered single-question survey about catnip response could potentially be cheap, while delivering an unbiased estimate from the general population uninterested in cats or catnip (particularly in Japan, given the Japanese anomaly in the meta-analysis), and incidentally measuring proportions that might also be of interest (such as international differences in cat ownership or familiarity with catnip). The disadvantages are that the cost advantage may be illusory as most respondents simply will not have a cat or will not have tried to use catnip (never mind any of the more obscure cat stimulants like silvervine), and the survey must be kept as simple as possible to keep respondents honest & costs down.

Given the screening question feature, a single-question catnip survey could be defined either as a two-stage survey with a screening question asking about cat ownership and then results of catnip use if any, or as a single-question asking about catnip with the set of responses no-cat/cat-but-never-catnip/cat-and-response/cat-and-no-response. If half of Americans have a cat, and half of cat-owners have tried catnip, then one would need 4 responses for every useful response, so it would cost $0.40/response. The screening question might be more efficient.

Pilot surveys

I experimented with two screening question GS surveys, Do you have a cat? and Have you ever given catnip to a cat?; after its automatic pilot experiment, GS priced responses to both at $3.00 each. I then experimented with two single-question GS surveys. The first GS survey was the straightforward single-question version to try to estimate what fraction of respondents would not have a cat & not have tried catnip, to see what the cost might be; the second reversed the responses to ask about catnip immunity rather than response, to try to reduce any demand bias:

  1. 28 July 2017 - 30 July 2017, 100 responses x $0.10 = $10, USA general population (raw responses)

    1. Have you ever given catnip to your cat?

    [randomized order]

    • No: I do not have a cat [n=65 / 62%]
    • No: I have a cat but have not tried catnip [n=12 / 16%]
    • Yes: but they did not respond to catnip [n=6 / 5%]
    • Yes: they responded to catnip [n=18 / 17%]
    24 useful responses, implying a 25% immunity rate.
  2. 29 July 2017 - 1 August 2017, 200 responses x $0.10 = $20, USA general population (raw responses):

    1. Have you ever given catnip to your cat?

    [randomized order]

    • No: I do not have a cat [n=136 / 72%]
    • No: I have a cat but have not tried catnip [n=19 / 9.9%]
    • Yes: and they were catnip-immune [n=11 / 4.7%]
    • Yes: but they were not immune [n=34 / 13.3%]
    45 useful responses, implying a 24% immunity rate.
  3. combined, n=300:

    • No cat: n=201 / 67%
    • Cat but no tries: n=31 / 10%
    • Cat and tried: n=69 / 23% ; result:

      • none/immune: n=17 / 24%
      • responder: n=52 / 75%

Pooling, about 1 in 7 or 15% (34+11100+200=34300=0.15\frac{34+11}{100+200}=\frac{34}{300}=0.15) of the responses were useful, so the true cost of one response was ~$0.67; so the screening question survey would cost 4.5x more than the single-question version. (Google Surveys sometimes errs on the side of collecting too many responses, but does not appear to charge you for the excess, so the cost-effectiveness will be a little bit better than appears.) Demand bias wise, the proportions are identical between the normal & negated responses, providing some evidence that respondents are not mindlessly responding. Combining the pilot survey samples and comparing with the earlier meta-analytic estimate of 63% response rate / 37% immunity observed in lab animals/research settings, the difference would be just statistically-significant when ignoring the meta-analytic uncertainty (binom.test(c(17, 52), p=0.37) ~> p=0.034, CI 0.15-0.365), so it may be the case that the cat owners are more likely to say their cat responds to catnip.

Full International Surveys

This difference, and the apparent success of the pilot survey, suggests the need for a larger survey sample to nail down the estimate more accurately, and exploit the ability to sample other countries (especially Japan). To sample all 9 non-US countries (omitting Italy & the Netherlands because of the unknown biases in sampling only Android smartphone users), I will need translations of the questions into: French, German, Japanese, Portuguese, & Spanish.

  1. Canada: 7-14 August 2017, 3000 responses x $0.10 = $300, Canada general population (raw responses)

    1. Have you ever given catnip to your cat? [n=3001]

    [randomized order]

    • No: I do not have a cat [n=1938 / 63%]
    • No: I have a cat but have not tried catnip [n=260 / 9%]
    • Yes: but they did not respond to catnip [n=139 / 5.3%]
    • Yes: they responded to catnip [n=664 / 23%]
    $=$17% immunity rate. Canadians appear more likely to have cats, try catnip, and receive a catnip response.
  2. Canada: 19-22 August 2017, 150 responses (raw responses); adaptive sampling followup:

    • no: 99 (62%)
    • non-trier: 12 (9.5%)
    • non-response: 8 (5.9%)
    • response: 31 (23%)
    831\frac{8}{31}=25.8% immunity rate.
  3. Australia: 7-10 August 2017, 3000 responses x $0.10 = $300, Australia general population (raw responses)

    1. Have you ever given catnip to your cat? [n=3000]

    [randomized order]

    • No: I do not have a cat [n=2134 / 71%]
    • No: I have a cat but have not tried catnip [n=446 / 14.3%]
    • Yes: but they did not respond to catnip [n=196 / 6.8%]
    • Yes: they responded to catnip [n=224 / 7.7%]
    $=$47% immunity rate. This is surprisingly high: almost twice the US/UK estimates even though one would expect that Australian rates would be similar as it was colonized by the English & presumably English cats. It raises the same question as the Japanese anomaly: are there large cross-national differences in catnip response rates, and if so, might they be genetic in origin, perhaps due to founder effects or genetic drift?
  4. Australia: 12-15 August 2017, 2000 responses x $0.10 = $200, Australia general population (raw responses); followup survey to confirm the 47% anomaly in the first Australian survey:

    1. Have you ever given catnip to your cat? [n=2001]

    [randomized order]

    • No: I do not have a cat [n=1438 / 71%]
    • No: I have a cat but have not tried catnip [n=293 / 15%]
    • Yes: but they did not respond to catnip [n=120 / 7.4%]
    • Yes: they responded to catnip [n=150 / 6.5%]
    $=$44%. Pooled: $=$46%.
  5. Australia: 22-25 August 2017, 500 responses x $0.10 = $50, Australia general population (raw responses); adaptive sampling followup:

    • non-owner: 361 (74%)
    • non-trier: 75 (16%)
    • responder: 29 (5%)
    • non-responder: 25 (4.9%)
    46% immunity rate, no surprise there.
  6. Australia: 29-TODO August 2017, 150 responses x $0.10 = $15, Australia general population

    • Non-owner: (%)
    • Non-trier: (%)
    • Immune: (%)
    • Responder: (%)
  7. UK: 7-10 August 2017, 3000 responses x $0.10 = $300, United Kingdom general population (raw responses)

    1. Have you ever given catnip to your cat? [n=3021]

    [randomized order]

    • No: I do not have a cat [n=2131 / 70%]
    • No: I have a cat but have not tried catnip [n=265 / 9%]
    • Yes: but they did not respond to catnip [n=162 / 5.5%]
    • Yes: they responded to catnip [n=463 / 15.6%]
    $=$25.92% immunity rate.
  8. USA: 9-11 August 2017, 2700 responses x $0.10 = $270, USA general population (raw responses)

    1. Have you ever given catnip to your cat?

    [randomized order]

    • No: I do not have a cat [n=1826 / 65%]
    • No: I have a cat but have not tried catnip [n=269 / 9.6%]
    • Yes: but they did not respond to catnip [n=151 / 5.1%]
    • Yes: they responded to catnip [n=563 / 20.5%]

    $=$22% immunity rate. Pooling with previous 2 USA pilot surveys for a combined n=3000:

    • No cat: n=2027 / 68%
    • Cat but no tries: n=300 / 10%
    • Cat and tried: n=783 / 26% ; result:

      • none/immune: n=168 / 21%
      • responder: n=615 / 79%
  9. Mexico: 11-13 August 2017, 3000 responses x $0.10 = $300, Mexican general population (raw responses); Spanish version provided by David Figuera:

    1. ¿Alguna vez le ha dado nébeda (catnip/menta gatuna) a su gato? [n=3011]
    • No: No tengo gato. [n=2245 / 74.2%]
    • No: Tengo gato pero no lo he probado. [n=594 / 19.8%]
    • Sí: Pero no le hizo efecto. [n=82 / 3.3%]
    • Sí: Le hizo efecto. [n=90 / 2.7%]
    $=$48% immunity rate. This is, like the first Australian sample, interestingly high but compromised by relatively small sample size: Mexicans apparently tend to own cats less and to be less likely to try catnip, so barely 5% of responses are useful.
  10. Mexico: 19-22 August 2017, 150 responses (raw responses); adaptive sampling followup:

    • Non-owner: 112 (75%)
    • Non-trier: 30 (20%)
    • Responder: 4 (2.5%)
    • Immune: 4 (2.2%)
    $=$50%.
  11. Spain: 11-14 August 2017, 3000 responses x $0.10 = $300, Spanish general population (raw responses)

    1. ¿Alguna vez le ha dado nébeda (catnip/menta gatuna) a su gato? [n=3000]
    • No: No tengo gato. [n=2203 / 73.4%]
    • No: Tengo gato pero no lo he probado. [n=607 / 21.2%]
    • Sí: Pero no le hizo efecto. [n=87 / 2.4%]
    • Sí: Le hizo efecto. [n=103 / 3%]
    $=$46% immunity rate. Similar to Mexico: low rates of cat ownership & catnip trying, high immunity rate.
  12. Spain: 19-22 August 2017 (raw responses); adaptive sampling followup:

    • Non-owner: 97 (68%)
    • Non-trier: 43 (24%)
    • Immune: 5 (4.7%)
    • Responder: 5 (3.7%)
    $=$50% immunity rate.
  13. Germany: 13-16 August 2017, 3000 responses x $0.10 = $300, German general population (raw responses); German version provided by r0kit, checked by FeepingCreature & gehmehgeh4:

    1. Haben Sie Ihrer Katze jemals Katzenminze/Catnip gegeben? [n=3009]
    • Ich habe keine Katze. [n=2093 / 71.3%]
    • Nein, bisher habe ich ihr keine Katzenminze/Catnip gegeben. [n=536 / 16.4%]
    • Ja, aber sie reagiert nicht darauf. [n=164 / 4.8%]
    • Ja, ich habe ihr Katzenminze/Catnip gegeben und sie reagiert darauf. [n=216 / 7.4%]
    $=$43% immunity rate. Somewhat intermediate Spain & UK.
  14. Japan: 22-25 August, 200 responses x $0.10 = $20, Japanese general population (raw responses); Japanese version provided by Juju Kurihara

    1. 自分のネコにキャットニップあげたことある? [n=203]

    [randomly reverse answer order; GS forbade full randomization]

    • いいえ。ネコは飼ってない。 [n=162 / 84%]
    • いいえ。ネコはいるけど、キャットニップをあげたことない。 [n=21 / 9%]
    • はい。でも反応がなかった。 [n=6 / 1.9%]
    • はい。反応した。 [n=14 / 4.7%]
    $=$30% immunity rate. Consistent with a low immunity rate but too small to rule out 10%/90%, requiring additional sampling.
  15. Japan: 24-28 August, 3800 x $0.10 = $380, Japanese general population (raw responses); modified Japanese version with a catnip synonym for clarity, larger sample size:

    1. 自分のネコにキャットニップ(イヌハッカ)あげたことある? [n=3828]

    [randomly reverse answer order]

    • いいえ。ネコは飼ってない。 [n=3158 / 83%]
    • いいえ。ネコはいるけど、キャットニップ(イヌハッカ)をあげたことない。 [n=383 / 10.4%]
    • はい。でも反応がなかった。 [n=82 / 2.4%]
    • はい。反応した。 [n=205 / 4.5%]
    $=$29% immunity rate; pooling: 88307\frac{88}{307}=29%. This is an immunity rate similar to the UK and not otherwise remarkable, so the Japanese anomaly has been falsified.
  16. Japan: 29-TODO August, 150 x $0.10 = $15, Japanese general population

    • Non-owner: (%)
    • Non-trier: (%)
    • Immune: (%)
    • Responder: (%)
  17. Brazil: 25-27 August, 3000 x $0.10 = $300, Brazilian general population (raw responses): Portuguese translation provided by Gladstone:

    1. Você alguma vez deu erva-dos-gatos para seu gato? [n=3045]

    [randomized order]

    • Não: Não tenho gato. [n=1951 / 64%]
    • Não: Tenho gato, mas nunca tentei dar. [n=781 / 26%]
    • Sim: Mas não fez efeito. [n=139 / 4.5%]
    • Sim: E fez efeito. [n=174 / 5.7%]

Overall results as a table:

Country Start End Total Non-owners Non-triers Triers Immune Responders Immunity rate $/response
USA 2017-07-28 2017-08-11 3110 2027 (68%) 300 (10%) 783 168 615 21% $0.38
UK 2017-08-07 2017-08-10 3021 2131 (70%) 265 (9%) 625 162 463 26% $0.48
Australia 2017-08-07 2017-08-25 5501 3933 (71%) 814 (15%) 744 341 403 46% $0.74
Canada 2017-08-07 2017-08-22 3151 2037 (65%) 272 (9%) 842 147 695 18% $0.37
Mexico 2017-08-11 2017-08-22 3161 2357 (75%) 624 (20%) 180 86 94 48% $1.75
Spain 2017-08-11 2017-08-22 3150 2300 (73%) 650 (21%) 200 92 108 46% $1.58
Germany 2017-08-13 2017-08-16 3009 2093 (71%) 536 (16%) 380 164 216 43% $0.79
Japan 2017-08-22 2017-08-28 4031 3320 (83%) 404 (10%) 307 88 219 29% $1.30
Brazil 2017-08-25 2017-08-28 3045 1951 (64%) 781 (26%) 313 139 174 44% $0.96
  • TODO: still need French translation
  • TODO: check when I’m done: does total cat ownership rate, or catnip trying rate, predict catnip response rate? if cats tend to not respond to catnip, people will bother less, and owning cats won’t be as fun

convenience Sampling

Collecting data by hand is very slow unless I make a major effort like getting permission from a local animal shelter to spend a day or two testing all their cats. But surveying cat owners could be a shortcut to getting a lot of data. (Only a small percentage will have gone beyond catnip, but that’s still a bigger sample than I have now.)

Prizes: 3 $50 Amazon gift cards Survey: Mechanical Turk; Facebook? can do advertising or a boosted post to https://www.facebook.com/CatsLoveCatnip/ can do MT by sending Turkers out to a copy of the Google Docs survey and giving them a completion code at the end

sample sizes: need >33 catnip respondents per country for detecting possible Japan effect limit MT to Australia/China/Taiwan/Korea/Japan (so 5*>33 = >165 total from MT) cat-owner qualification? avoid overpaying by the <10 HIT strategy https://docs.google.com/presentation/d/1Y_lvecsOefCfkXkkrdFtW1GH-NkoZidKzKXqFVJaLH4/edit#slide=id.ga3bec7adb_1_6 so at least 19 HITs MT calculator: https://morninj.github.io/Mechanical-Turk-Cost-Calculator/ 165 responses at $0.50 reward (for high quality responses) + MT’s 20% commission = $108

preview: https://docs.google.com/forms/d/e/1FAIpQLSeT3GIg-pSwzDFAfNaqE-MzfJEtD0HghN_Vma68OZJtz1Pztg/viewform launched 29 September 2016:

demographics:

  • gender
  • age
  • country

For 5 cats, ask:

  • sex (M/F)
  • spayed/neutered
  • breed: various
  • fur color; used cat color list from The Relationship Between Coat Color and Aggressive Behaviors in the Domestic Cat, Stelow et al 2016:

    • black
    • black-and-white
    • calico
    • color points
    • gray
    • gray-and-white
    • Tabby (black, brown, and gray)
    • Tabby (orange, cream, and buff)
    • tortoiseshell
    • white
    • Torbie
    • other
    • personality 1-5: The personality and emotional factors were found to be the most important: withdrawn cats react poorly while friendly, outgoing cats react best
  • age at first administration
  • stimulants, binary: catnip response, valerian response, honeysuckle response, silvervine response, cat thyme

Catnip:

  • has the person ever consumed catnip in the form of: tea / leaves or an herb / roots / smoked / poultice
  • if so, what was the purpose? relaxation / stimulation / euphoria or intoxication / hallucination or visual distortion / dreams / insomnia / stomach aches / mosquito repellent / headaches / colds or flu or fever / hives / arthritis / increasing urination / treatment of worms / hemorrhoids / other
  • efficacy: 1-5

Post-launch edits:

  • removed free response from gender question in demographic section due to abuse
  • added age constraints to the human and cat age questions due to abuse
  • edited introduction to rephrase it as cat non-responses & began advertising as a catnip non-response survey: the first 49 respondents claimed that 45 of 47 cats (cat #1 responses) responded to catnip, which is grossly discrepant from the 62% estimate - it’s imprecise, but there’s no way the true catnip rate is 96%! Further, the catnip non-response rate goes up in the additional cat entries. All this indicates either a bias in respondent recollections (they only remember the cats which do respond, or they provide a responding cat’s data first and then don’t include all the rest they know of) or a response bias to experimenter demand (inferring that I want to hear only about cats which do respond). The response rates for the other substances like silvervine are more reasonable thus far (although with far smaller sample sizes) and the small sample size is what I expected because those substances are much rarer, so this may not be a general acquiescence bias (because you would expect many more people to be claiming their cat respond to all of catnip/silvervine/valerian/thyme). If it’s a recollection bias, I don’t know of anyway to correct that afterwards or change the survey to eliminate it, so surveys on this topic might be futile.

Known cat stimulants

Catnip, Valerian, Honeysuckle and other cat-attractant plants, Hartwell 2008

  • Nepeta cataria (catnip) https://en.wikipedia.org/wiki/Catnip#Effect_on_cats : active ingredient nepetalactone
  • Lonicera Tatarica (Tatarian Honeysuckle) https://en.wikipedia.org/wiki/Lonicera_tatarica : nepetalactone TODO: that can’t be the only ingredient if catnip doesn’t work on Percy and honeysuckle does?
  • Valerian:

    • Valerian officinalis https://en.wikipedia.org/wiki/Valerian_%28herb%29#Effect_on_other_organisms : active ingredient actinidine
    • Valeriana celtica https://en.wikipedia.org/wiki/Valeriana_celtica : active ingredients nepetalactone (Bicchi et al 1983)
  • Actinidia Polygama (Japanese Catnip or Matatabi or Silvervine or silver vine or cat powder) https://en.wikipedia.org/wiki/Actinidia_polygama#Pets and Actinidia macrosperma : active ingredient Actinidine and dihydroactinidiolide

  • Acalypha indica (Indian acalypha/Indian nettle) active ingredient TODO
  • cat thyme (teucrium marum) : active ingredient TODO
  • Buckbean (Menyanthes trifoliata): mitsugashiwalactone

  • Northern groundcone (Boschniakia rossica): boschniakine and boschnialactone
  • Yellowbells (Tecoma stans): boschniakine and actinidine
  • Trumpet Creeper (Campsis radicans): may boschniakine
  • Guelder Rose (Viburnum opulus, sometimes called Cranberry Bush and most commonly found in cultivation as the Snowball Tree)
  • the perennial Dittany of Crete (Origanum dictamnus/Hop Marjoram)
  • the spring-flowering annual Baby Blue-eyes (Nemophila menziesii)
  • the Zimbabwean plant Zinziba (Lippia javanica aka Verbena javanica).

Speculative:

  • there are anecdotal reports that some cats are attracted to olive products: olive wood, olive oil, and olive fruits themselves

Local cat experiments

The catnip literature has a good sample of what the catnip frequency response is, but it’s clear that a lot of cats are immune and their owners need to use one of the many substitutes. The literature does not seem to have much, if any, coverage of all the main alternatives, so owners don’t know what to try after catnip. There’s also no information on any correlations between responses: does catnip immunity cause susceptibility to a different stimulant? Are some cats more response to stimulants in general? If we know a cat is immune to, say, catnip and valerian, can we better predict what a owner should try next?

Since cats are easy to come by and the stimulants are cheap, I thought it might be interesting to assemble a collection of stimulants and systematically try them out on each cat I should happen to meet. Specifically, I’ll look at the most common and easy to come by stimulants, since it’s not useful to discover that some extinct African flower is super-potent:

  • Catnip (obviously)
  • Tatarian honeysuckle (one of the most frequently mentioned alternatives)
  • Valerian (not as commonly mentioned, but very cheap due to its popularity as a herbal supplement for sleep & anxiety)
  • Silvervine (rarer & more expensive but still accessible, and reputedly quite effective)
  • Cat thyme
  • Olive oil/wood (cats’ reacting to olive-related things is an obscure & anecdotal claim; but both are easy to get and included for the sake of comprehensiveness)

Purchasing

For each stimulant, one clean never-used feather was put into a clean jar along with a good helping of the relevant stimulant, and shaken and left to sit. The olive wood would not fit into a jar, so I sawed it in half, saved the sawdust, and put the two wood blocks, sawdust, and feather into a little ziplock bag. The wand lets the feathers be added or removed easily so I can test each feather one at a time by switching feathers, playing with the cat for 5 minutes, and waiting for any response or indication of interest beyond regular chase. This works well, aside from the silvervine powder where inevitably some of the powder falls off or blows out of the container, so it’s best to do the silvervine last.

Efficacy

Cat Sex Age Fixed Color Kind Catnip Valerian Honeysuckle Silvervine Olive wood Olive oil
Oolong M 1.5 1 black Domestic long 0 1 0 1 0 0
Percy M 3 1 brown Tabby 0 0 1 1 0 0
Kiki M 2 1 grey Tabby 1
Stormy M 10 1 grey Russian Blue
Sara F 10 1 brown Tabby 1
Whitey F 5 1 white Domestic short 1
Sassy F 12 1 orange Domestic short 0 0 0 1 0 0

Description of effects:

  • Catnip: too well known to need much description. Catnip responder cats become hyper, uninhibited, playful, interested in the catnip toy.

    Testing it in my 2 non-responder cats in December 2014, Percy and Oolong initially sniffed it for a few seconds and then ignored it afterwards, regardless of whether I shook a quarter kilogram of catnip leaf under their nose or let the open jar sit next to them. Likewise, attempts to interest them in the catnip extract spray in November 2015 also failed. In October 2016, I purchased ~100g of dried catnip leaves at a local Renaissance festival to have a backup for the catnip extract spray and perhaps try out catnip tea myself (and if I can’t find any use for it, I can always experiment with home steam distillation of catnip oil from catnip leaves!). Retesting Percy/Oolong with the new leaves, Percy remained uninterested but this time Oolong was interested in the plastic bag containing the catnip leaves, trying to claw and chew it, so I put ~10g into a little cotton bag and give it to him, which he began to chew and claw and even laid down to hold it in his paws and kick at it - however, despite resembling the catnip response, he was not hyper, and he did not repeat the reaction on 4 subsequent occasions I tried to interest him in the catnip bag (spraying it with the catnip extract did not help). This makes me wonder if catnip response might have some ultra-long tolerance like some drugs in humans which take weeks or months for tolerance to be restored, or if the catnip response is not bimodal after all (catnip being an dominant gene would certainly seem like it almost has to produce a bimodal/binary trait, but we have only Todd 1962 as evidence for it being dominant, and he used a single small breeding colony and no one has ever followed up & tried to replicate it that I know of). Perhaps the issue of measurement error has been underestimated in past studies and many cats will only display a catnip response occasionally or under ideal conditions, and so the estimates of catnip immunity are heavily biased upwards?
  • Valerian: on Oolong, produces an interesting mix of passiveness and possessiveness - after onset and a long Flehmen response, he mostly lays on the floor passively, occasionally pulling himself across it towards the toy, but generally making little effort to hunt; if the toy comes within reach, though, then he seizes it energetically and abruptly begins clasping it to his face, chewing it, curling up around it and kicking at the toy with his hind legs. Eventually he can be coaxed into playing chase normally, with his usual level of competence and interest. Actively repelled Sassy.
  • Honeysuckle: on Percy, had a general stimulant effect; despite being fat, lazy, and usually entirely uninterested in chase, he will make an effort to play after sniffing a honeysuckle-impregnated toy. This stimulating effect does not produce the un-inhibition of catnip, and seems fairly mild. Actively repelled Sassy.
  • Silvervine:

    1. on Oolong, a general stimulating effect with considerable interest in playing chase or watching the toy, but curiously, he makes many fewer attempts at attacking or chasing and when he does, he is distinctly slower (and thus, less effective) than usual. His coordination is fine - he’s not clumsy or falling over - but he’s just not as effective (in a way hard to pin down specifically).
    2. On Percy, likewise a general stimulating effect but far more effective than the honeysuckle in inducing chase play, with him even trying out jumps when coaxed appropriately; he may be less effective, like Oolong, but he plays chase so little that it’s impossible for me to make any comparisons with his normal chase behavior.
    3. Sassy: similar. More time on back while playing chase than usual.
  • Olive oil/wood: all cats showed brief interest in smell, then ignored entirely.

Appendix

Breeding cats to increase frequency of catnip response

Given the benefits of catnip for behavioral enrichment, it’s unfortunate that cat breeders have not used a little selection pressure to make catnip response universal. There are 3 major scenarios for catnip breeding:

  1. Mendelian autosomal dominant: marker-assisted genetic selection: 1 generation to fixation

    A genetic test for catnip response is not available as of July 2017, since the allele has not even been localized to any chromosomes, much less identified, and given the small niche of cat genetics research, large-scale GWAS would almost certainly be very expensive (eg the 99 Lives Cat Genome Sequencing Initiative requires, as of July 2017, ~$7000 to sequence 1 cat genome, which is >7x more than it would cost to sequence a human genome via Veritas, and they have sequenced 53 cats). As catnip response appears somewhat similar across the felids, a successful find of a catnip gene in another species might make it much easier to find it in cats, reducing the sample size requirement.

    Fortunately, cats are otherwise not expensive to test, as a number of successful existing feline saliva-based genetic tests indicate (Lyons 2009, Lyons 2010, Lyons 2012): as of August 2017, the UC Davis Veterinary Genetics Laboratory offers coat color, ancestry, parentage, blood group, genetic diseases (Progressive Retinal Atrophy/PRA-PK Deficiency, Glycogen Storage Disease Type IV, GM1 & GM2 Korat Gangliosidosis, Polycystic Kidney Disease etc) for prices ranging $40-$120; the UK Langford Vets (University of Bristol) offers a similar list of tests for a flat £35.40/$47; the US CFA’s CatDNAtest.org offers a more limited selection but again all for $45. The Morris Animal Foundation sponsored development of an Illumina feline SNP array (the Illumina Infinium iSelect 63K Cat DNA Array) starting around 2011 (with unfortunately a limited selection of only 70k SNPs as opposed to the 500k+ standard in human SNP arrays), and as of 2014, was offering it for$100 to researchers. Hence, it is plausible that after discovering the catnip allele, the cost of testing a single cat might be <$50.

    Once a genetic test does become available at sufficiently low cost, it offers by far the fastest and most powerful selection method. Given a genetic test for the catnip allele (as a common variant, it is likely easily tested for), breeding could be conducted in 1 generation: test the population of catnip responders, and breed only the homozygous dominant cats (ie the cats with 2 copies), who make will make up ~0.442=0.190.44^2=0.19 of the population. All offspring will then also be homozygous dominant & thus catnip responders, and catnip response rates will be 100% barring any further introduction from outside the breeding population of recessive carriers. This sort of single-generation selection would be doable at fairly modest total cost, <$100k (eg start with 1000 cats, test for catnip response; about half will respond, and testing each at $50 would cost ~$25k, of which about half will be homozygous rather than heterozygous, giving a large founding population of ~250 cats all of whose descendants will be catnip responders).
  2. Mendelian autosomal dominant: phenotypic selection: ~9 generations to near-universality (>=99%)

    Assuming there is no genetic test, one can instead breed based on the observed phenotype of catnip response; this will be slower and less efficient because heterozygous and homozygous cannot be distinguished.

    But if Todd is correct about the allele being dominant, catnip response is still easily bred, since selection against a recessive proceeds fast when starting from a high frequency. Under the Mendelian autosomal dominant theory, if the catnip recessive is present at q=0.56q=0.56 per Todd 1962, and we would like to decrease the rate of catnip resistance in a breed to <=1%, then we need to decrease the frequency to 0.01=0.10\sqrt{0.01}=0.10. If we breed only catnip responders (which fortunately are the majority already), this corresponds to a selection intensity s=1, and the number of generations has the simple equation (more general treatments: 1, 2) n=1qn1q0n = \frac{1}{q_n} - \frac{1}{q_0} or 10.0110.56=8.2\frac{1}{\sqrt{0.01}} - \frac{1}{0.56} = 8.2 generations. Cats mature reproductively at 18 months & pregnancy is around 2 months, so 8.2 generations would take ~14 years ((18+2)8.212\frac{(18+2) \cdot 8.2}{12}). So a phenotypic catnip breeding program need not take an infeasible amount of time based on Todd 1962’s estimates.
  3. polygenic liability-threshold: phenotypic selection: ~25 generations to near-universality

    Alternatively, if Todd is incorrect about catnip being Mendelian dominant controlled trait, it still appears to run in families & species and thus have a genetic basis, so, since the trait appears binary, the true genetic architecture might then be a liability threshold trait. Phenotypic selection in the liability threshold trait depends on the population frequency of the trait (which defines the unobserved latent threshold’s starting point, and efficacy slows at the extremes of 0 and 1) and on the heritability of the trait (which defines the genetic response to each generation of selection; unfortunately, unknown for cats). Large-scale human twin studies such as Polderman et al 2015 turn in a grand mean across all studied human traits of ~50% heritability, so a priori one would expect a reasonably high heritability, and Villani 2011’s large catnip results indicate, considering the measurement error, a heritability of at least 0.7, so we will consider that scenario:

    threshold_select <- function(fraction_0, heritability) {
        fraction_probit_0 = qnorm(fraction_0)
        ## threshold for not manifesting schizophrenia:
        s_0 = dnorm(fraction_probit_0) / fraction_0
        ## new rate of schizophrenia after one selection where 100% of schizophrenics never reproduce:
        fraction_probit_1 = fraction_probit_0 + heritability * s_0
        fraction_1 = pnorm(fraction_probit_1)
        ## how much did we reduce schizophrenia in percentage terms?
        print(paste0("Start: population fraction: ", fraction_0, "; liability threshold: ", fraction_probit_0, "; Selection intensity: ", s_0))
        print(paste0("End: liability threshold: ", fraction_probit_1, "; population fraction: ", fraction_1, "; Total population reduction: ",
            fraction_0 - fraction_1, "; Percentage reduction: ", (1-((1-fraction_1) / (1-fraction_0)))*100))
        return(fraction_1)
        }
    heritability <- 0.70; frac <- 0.61
    threshold_select(frac, heritability)
    # ...
    # [1] 0.7641164429
    generations <- 0;
    while (frac < 0.99) { frac <- threshold_select(frac, heritability); generations <- generations+1 }
    # ...
    generations
    # [1] 25

    Progress, while initially swift (eg getting to 95% only takes ~7 generations), slows down rapidly towards the target of 99% response rates. The selection, especially towards the end, could be improved by taking into account genetic information such as pedigree or GWAS-based polygenic scores in order to select against cats with a high liability for catnip resistance but who do not display the phenotype; modeling that, however, would be difficult without more data.

Does initial data support one-and-only one response per cat?

When I tested the first two cats with the 3 stimulants I had, they both responded to exactly 1 of the 3 stimulants: Oolong ignored catnip & honeysuckle while responding to Valerian, and Percy ignored catnip & Valerian to respond to honeysuckle. Honeysuckle and catnip supposedly have the same active ingredient while Valerian differs, so this is an odd pattern. Could there be something at play like a single mechanism which responds to only one chemical and so every cat responds to one-and-only-one-kind (OO) of stimulant? The existing literature doesn’t seem to describe many (any?) tests of multiple stimulants on the same cat… Is this much evidence?

Assume binary responses, that the default or null hypothesis theory here is that each stimulant has a 50% chance of working (I only know the population frequency for catnip, which is somewhat close to 50%), and they are independent; then the exact probability of drawing 1 of 3 is a binomial with 1 success of 3 trials with 50% chance of success or ~0.375 (through some combinatorics and can be checked by simulation), and then the probability of two such events is their conjunction of 0.3752=0.141. This 0.141 can be interpreted as a p-value (the frequency with which the null hypothesis generates this particular set of stimulant responses if we sampled two cats). If we didn’t believe the probability of 0.375 pulled out of combinatorics, we could do a simulation-based hypothesis test by twice generating a possible value from a binomial distribution and seeing in how many simulations both datapoints (cats) yielded exactly 1 of 3, which matches our 0.141 p-value:

simulates <- 100000
matches <- sum(rbinom(simulates, 3, p=0.5)==1 & rbinom(simulates, 3, p=0.5)==1)
matches / simulates
# [1] 0.14012

OK, so we know under a plausible alternative model, our data from 2 cats occurs fairly often, but what does that mean? A Bayesian model-comparison would be more meaningful.

In this case, we compare the binomial model with a model in which each cat responds to OO stimulant. We can use ABC to continue the theme. One might object, don’t we already know the result of an ABC? If we write a generative model for the OO theory, it will yield the data every time, and we already know the binomial theory yields the data only 14% of the time, so we get a ratio of 10.14=7\frac{1}{0.14}=7, which is an interesting level of evidence (keeping in mind that we must have a low prior on the idea of only one stimulant, since it didn’t come to mind before we saw the data).

I would say that’s not quite right, because here we have the twist that we still don’t have the silvervine data - if it’s true cats respond to only one stimulant, then silvervine could have been either cat’s special stimulant and so they would appear to have no responses in our current data. The OO theory is that it would be out of the major stimulants, which are catnip, honeysuckle, Valerian and silvervine. So we need to account for the penalty that it predicts data other than we saw (like data in which neither cat responds to any of the 3 stimulants), which makes a BF of 7 to be an overestimate.

Here for ABC we could simulate this, but again the exact answer is easy: If there’s 4 stimulants and we sample 3, and there is exactly 1 hit, the chance of observing it is 34\frac{3}{4} (since the chance of not observing it, of it lurking in the last stimulant, #4, is 14\frac{1}{4} and 114=341-\frac{1}{4}=\frac{3}{4}); and the probability of observing this twice is 342\frac{3}{4}^2 or 0.5625. So the probability of the data under the null hypothesis is 0.141 and the probability under this alternative OO model is 0.5625, giving a ratio of 4 in favor of it (substantially less than the 7 before).

Since I find the OO theory to be ad hoc, based on looking at the data rather than modeling, and also weird - why would Percy react to honeysuckle and not catnip given that the active ingredient supposedly is nepetalactone in both? - I would give it a negligible prior probability like 1%, in which case a BF of 4 would increase only to around 4%.


In any case, after doing those calculations, I was able to try some more potential stimulants: silvervine, olive wood, and olive oil. Under the OO theory, since we had already seen a response from both cats (for valerian & honeysuckle), we should see no further responses; but unsurprisingly, I did, as both of them responded to silvervine (though neither olive). So we can probably reject the theory.


  1. For example, Todd 1962’s genetic analysis was based on just 34 cats of which 7 cats’ response were unknown; as far as I can tell, another pedigree analysis has never been done since then, much less extended to newer genetic methods like linkage or GWASes, although it continues to be cited as the justification for treating catnip response as an autosomal dominant genetic trait.

  2. An example of founder effects in cat populations are the feral cats of Kerguelen Islands, whose fur colors are >95% black (or black with white spots): Dreux 1970, Todd 1971, Dreux 1974, Pascal 1994, Pontier et al 2004

  3. If catnip sensitivity frequency was 70% in 1950, and for the sake of argument we accepted that the proportion had decreased ~0.0024 per year and was 54.16% in 2016, and we also accept Todd 1962’s analysis that catnip sensitivity is a dominant gene, then that implies that the frequency of the gene in 1950 was 0.7=p2+2pq0.7 = p^2 + 2pq and so in 1950 p=0.45; similarly, 0.516=p2+2pq0.516 = p^2 + 2pq, and so in 2016 p=0.30, so the gene frequency would’ve fallen by 0.15 from 1950 to 2016. If a domestic or wild cat generation is 3 years, then that 66 years is >22 cat generations, and dividing a total fall of 15% over that period, the gene frequency would be falling ~0.68% per generation.

    While this could be caused by fairly weak selection against catnip responders, it’s hard to imagine what could cause that selection or how it is consistent with the continued existence of catnip responders given that: catnip has no known ill effects on cats, is rarely found in the wild, most cats are never exposed, those cats which are exposed at all tend to be pets which are spayed/neutered, no solid sex or breed correlation has been observed, and the existence of catnip responders has been documented for centuries (according to Tucker & Tucker 1988, John Ray, died 1705, was the first, and Gary Lockhart attributes the first mention of cats/catnip to Al-Biruni, died 1048).

  4. One interesting issue with running the German version of the survey in Google Surveys was dealing with a bug in Google Translate (GT). I noticed while examining the suggested translations in GT that it was mistakenly translating both I have a cat and I do not have a cat to the same Germany sentence, Ich habe eine Katze, which translated back as the single sentence I have a cat (failing the roundtrip criteria of identity, that En==En(De(En))); eine/Keine are the positive & negative German versions of have, so it was a major error. I thought nothing of it until GS refused to run the Germany survey, warning me that it needed editing so all respondents could answer the question, and suggesting that the response option I have a cat should be changed or a NA response added. Apparently GS automatically translates the non-English surveys back into English & checks that the responses are mutually exhaustive, so the first response was erroneously translated into English as I have a cat (rather than I do not have a cat) and the set of responses was indeed malformed (as all the responses then presume having a cat). After submitting a correction to GT & explaining this to the GS help desk, they restarted the survey within a few minutes.